feat(P0): 完成阶段一全部需求 — 混合检索/内容安全/用户反馈/FAQ精准匹配

P0-001 混合检索+重排序引擎:
- HybridSearchService 支持 VECTOR/KEYWORD/HYBRID 三种检索模式
- RrfFusion RRF 融合算法 (k=60)
- RerankerService 支持 DashScope + OpenAI 兼容多提供商
- vector_store 新增 content_tsvector 列 + GIN 索引 + 触发器
- DocSearch.js 增加检索模式下拉选择

P0-004 内容安全过滤:
- ContentSafetyService DFA 字典树引擎（volatile + copy-on-write 线程安全）
- ContentSafetyAdvisor BaseAdvisor 实现，输入/输出双向审核
- SensitiveWordService 敏感词 CRUD + 批量导入 + 字典树热加载
- SensitiveWordManager.js 前端管理组件

P0-002 用户反馈系统:
- MessageFeedbackService upsert 语义反馈提交 + 统计 API
- ChatPanel.js 添加 👍/👎 反馈按钮
- Chat SDK handleFeedback() 对接后端 POST /feedback
- ConversationService 导出集成反馈信息

P0-003 意图识别+FAQ精准匹配:
- IntentRouter LLM 意图分类（FAQ/RAG/CHITCHAT）
- FaqMatchEngine 三级匹配（精确→关键词→向量语义，阈值 0.85）
- FaqService FAQ CRUD + 异步向量化
- FaqManager.js 前端管理（CRUD + 批量导入/导出 + 启用/禁用）

共享变更:
- DatabaseInitConfig 新增 5 张表 + 全文检索初始化
- api.js 新增反馈/敏感词/FAQ API 封装
- app.js 注册 SensitiveWordManager + FaqManager 组件
- application.yml 新增 knowledge.faq.semantic-threshold 配置

CLAUDE.md

@@ -123,6 +123,40 @@ AI 智能客服系统，基于 Spring AI Alibaba + 通义千问 + PGVector，支
 - 文档管理: `/document/*`（`DocumentController`）
 - 批量操作: `/document/batch/*`（`DocumentController`，用 POST 避免 DELETE+RequestBody 路径冲突）
 - 分类管理: `/category/*`（`DocumentController`）
+- 消息反馈: `/feedback/*`（`MessageFeedbackController`）
+- 敏感词管理: `/sensitive-word/*`（`SensitiveWordController`）
+- FAQ 管理: `/faq/*`（`FaqController`）
+
+## P0 阶段新增功能
+
+### 内容安全过滤（P0-004）
+- **DFA 引擎**: `ContentSafetyService` 使用字典树匹配敏感词，`volatile` + copy-on-write 保证线程安全热加载
+- **ContentSafetyAdvisor**: 实现 `BaseAdvisor`，`getOrder()` 返回 `HIGHEST_PRECEDENCE`（Advisor 链最外层），before 阶段检查用户输入、after 阶段检查 AI 输出
+- **敏感词级别**: level=1 仅脱敏（MASK），level=2 拦截（BLOCK）返回安全提示
+- **审计日志**: `content_audit_log` 表记录所有违规事件，不删除
+- **前端**: `SensitiveWordManager.js` 在系统设置 Tab，支持 CRUD + 批量导入 + 审计日志查看
+
+### 用户反馈系统（P0-002）
+- **反馈实体**: `MessageFeedback`，按 `message_id` 唯一索引，重复提交覆盖（upsert 语义）
+- **反馈类型**: THUMBS_UP（有帮助）/ THUMBS_DOWN（没帮助），点踩可选原因分类 + 自由文本
+- **ChatPanel**: 每条 AI 回复下方有 👍/👎 按钮，点击调用 `POST /feedback`
+- **Chat SDK**: `handleFeedback()` 已连接后端 API，同时保留 localStorage 作为乐观 UI 缓存
+- **会话导出**: `ConversationService.exportConversation()` 导出的 TXT 中包含反馈信息
+
+### 意图识别 + FAQ 精准匹配（P0-003）
+- **IntentRouter**: LLM 单次调用做意图分类（FAQ/RAG/CHITCHAT），解析失败降级为 RAG
+- **FaqMatchEngine**: 三级匹配策略 — 精确匹配 → 关键词匹配 → 向量语义匹配（阈值 `knowledge.faq.semantic-threshold`，默认 0.85）
+- **FAQ 向量化**: 复用现有 `DynamicEmbeddingModel`，向量存入 `faq_embedding` 表，新增/修改 FAQ 时异步计算
+- **similar_questions 字段**: 使用 String 类型存储 JSON 数组字符串（PostgresJsonTypeHandler 期望对象格式，故不用 typeHandler）
+- **前端**: `FaqManager.js` 在知识库文档管理 Tab，支持 CRUD + 批量 JSON 导入/导出 + 启用/禁用
+
+### 混合检索 + 重排序引擎（P0-001）
+- **SearchMode**: 枚举 VECTOR（默认）/ KEYWORD / HYBRID，向后兼容
+- **HybridSearchService**: 多模式检索核心，KEYWORD 使用 PostgreSQL `tsvector` 全文检索，HYBRID 使用双路检索 + RRF 融合
+- **RrfFusion**: RRF 融合算法 `score = Σ 1/(k + rank_i)`，k=60
+- **RerankerService**: 支持 DashScope + OpenAI 兼容提供商，通过 `ai_model_config` 表 RERANK 类型配置，超时 3s 自动 fallback
+- **vector_store 全文检索**: 新增 `content_tsvector` 列 + GIN 索引 + PostgreSQL 触发器自动维护
+- **前端**: `DocSearch.js` 增加检索模式下拉选择（向量/关键词/混合），结果标注来源模式
 
 ## 已知 TODO
 

client/dist/chatbot-sdk.js

@@ -755,6 +755,36 @@ var ChatbotSDK = (function () {
             return [];
         }
     }
+    // ==================== P0-002: 消息反馈 ====================
+    /**
+     * 提交消息反馈（点赞/点踩）
+     */
+    async function submitFeedbackApi(messageId, feedbackType) {
+        if (!currentConfig)
+            return false;
+        const url = buildUrl('/feedback');
+        try {
+            const response = await safeFetch(url, {
+                method: 'POST',
+                headers: { 'Content-Type': 'application/json' },
+                body: JSON.stringify({
+                    messageId: String(messageId),
+                    conversationId: currentConfig.chatId,
+                    feedbackType,
+                }),
+            });
+            if (!response.ok) {
+                logger.error(`反馈提交失败 status=${response.status}`);
+                return false;
+            }
+            const json = await response.json();
+            return json.success || false;
+        }
+        catch (err) {
+            logger.error('反馈提交异常', err);
+            return false;
+        }
+    }
     /**
      * 获取会话列表
      */
@@ -3654,8 +3684,7 @@ var ChatbotSDK = (function () {
     // ==================== 消息反馈（👍 / 👎） ====================
     /**
      * 处理消息反馈：切换 AI 消息的点赞/点踩状态
-     * 当前仅前端状态持久化（存入 messages 数组 + localStorage），
-     * 预留后端对接位：TODO 接口 POST /conversation/message/feedback
+     * 前端状态持久化（存入 messages 数组 + localStorage）+ 调用后端 API 记录反馈
      */
     function handleFeedback(msgId, value) {
         if (!messagesContainer$1)
@@ -3673,7 +3702,21 @@ var ChatbotSDK = (function () {
         // 持久化（localStorage）
         if (config$1)
             saveMessages(config$1.integrateId, messages);
-        logger.info(`消息反馈 msgId=${msgId} value=${newValue || 'cleared'}`);
+        // 调用后端 API 记录反馈
+        if (newValue) {
+            const feedbackType = newValue === 'up' ? 'THUMBS_UP' : 'THUMBS_DOWN';
+            submitFeedbackApi(String(msgId), feedbackType).then(success => {
+                if (success) {
+                    logger.info(`消息反馈已提交 msgId=${msgId} value=${newValue}`);
+                }
+                else {
+                    logger.warn(`消息反馈提交失败 msgId=${msgId}（本地状态已更新）`);
+                }
+            });
+        }
+        else {
+            logger.info(`消息反馈已取消 msgId=${msgId}`);
+        }
     }
     function autoResizeInput() {
         if (!inputEl$1)

client/src/api.ts

@@ -376,6 +376,39 @@ export async function fetchRagSources(message: string, categoryId?: number): Pro
   }
 }
 
+// ==================== P0-002: 消息反馈 ====================
+
+/**
+ * 提交消息反馈（点赞/点踩）
+ */
+export async function submitFeedbackApi(
+  messageId: string,
+  feedbackType: 'THUMBS_UP' | 'THUMBS_DOWN'
+): Promise<boolean> {
+  if (!currentConfig) return false;
+  const url = buildUrl('/feedback');
+  try {
+    const response = await safeFetch(url, {
+      method: 'POST',
+      headers: { 'Content-Type': 'application/json' },
+      body: JSON.stringify({
+        messageId: String(messageId),
+        conversationId: currentConfig.chatId,
+        feedbackType,
+      }),
+    });
+    if (!response.ok) {
+      logger.error(`反馈提交失败 status=${response.status}`);
+      return false;
+    }
+    const json: ApiResponse = await response.json();
+    return json.success || false;
+  } catch (err) {
+    logger.error('反馈提交异常', err);
+    return false;
+  }
+}
+
 // ==================== P2: 会话管理 + chatId 初始化 ====================
 
 /** 会话列表项 */

client/src/chat.ts

@@ -20,6 +20,7 @@ import {
   updateChatId,
   getChatId,
   saveCachedChatId,
+  submitFeedbackApi,
   CskError,
 } from './api';
 import {
@@ -342,8 +343,7 @@ function announceMessage(text: string): void {
 
 /**
  * 处理消息反馈：切换 AI 消息的点赞/点踩状态
- * 当前仅前端状态持久化（存入 messages 数组 + localStorage），
- * 预留后端对接位：TODO 接口 POST /conversation/message/feedback
+ * 前端状态持久化（存入 messages 数组 + localStorage）+ 调用后端 API 记录反馈
  */
 export function handleFeedback(msgId: string, value: 'up' | 'down'): void {
   if (!messagesContainer) return;
@@ -362,7 +362,19 @@ export function handleFeedback(msgId: string, value: 'up' | 'down'): void {
   // 持久化（localStorage）
   if (config) saveMessages(config.integrateId, messages);
 
-  logger.info(`消息反馈 msgId=${msgId} value=${newValue || 'cleared'}`);
+  // 调用后端 API 记录反馈
+  if (newValue) {
+    const feedbackType = newValue === 'up' ? 'THUMBS_UP' : 'THUMBS_DOWN';
+    submitFeedbackApi(String(msgId), feedbackType).then(success => {
+      if (success) {
+        logger.info(`消息反馈已提交 msgId=${msgId} value=${newValue}`);
+      } else {
+        logger.warn(`消息反馈提交失败 msgId=${msgId}（本地状态已更新）`);
+      }
+    });
+  } else {
+    logger.info(`消息反馈已取消 msgId=${msgId}`);
+  }
 }
 function autoResizeInput(): void {
   if (!inputEl) return;

src/main/java/com/wok/supportbot/advisor/ContentSafetyAdvisor.java

@@ -0,0 +1,228 @@
+package com.wok.supportbot.advisor;
+
+import com.wok.supportbot.dao.ContentAuditLogMapper;
+import com.wok.supportbot.entity.ContentAuditLog;
+import com.wok.supportbot.service.ContentSafetyService;
+import lombok.extern.slf4j.Slf4j;
+import org.springframework.ai.chat.client.ChatClientRequest;
+import org.springframework.ai.chat.client.ChatClientResponse;
+import org.springframework.ai.chat.client.advisor.api.AdvisorChain;
+import org.springframework.ai.chat.client.advisor.api.BaseAdvisor;
+import org.springframework.ai.chat.messages.AssistantMessage;
+import org.springframework.ai.chat.messages.Message;
+import org.springframework.ai.chat.messages.UserMessage;
+import org.springframework.ai.chat.model.ChatResponse;
+import org.springframework.ai.chat.model.Generation;
+import org.springframework.ai.chat.prompt.Prompt;
+import org.springframework.beans.factory.annotation.Autowired;
+import org.springframework.core.Ordered;
+import org.springframework.stereotype.Component;
+
+import java.util.*;
+import java.util.stream.Collectors;
+
+/**
+ * 内容安全过滤 Advisor
+ * <p>
+ * 在对话链路的最外层拦截，对用户输入和 AI 输出进行敏感词检测与脱敏/拦截处理。
+ * 使用 Ordered.HIGHEST_PRECEDENCE 确保最先执行 before、最后执行 after。
+ */
+@Slf4j
+@Component
+public class ContentSafetyAdvisor implements BaseAdvisor {
+
+    /** 安全提示文本 —— 当用户消息被 BLOCK 时替换原始内容 */
+    private static final String SAFE_USER_PROMPT = "系统检测到不当内容";
+
+    /** 安全提示文本 —— 当 AI 输出被拦截时替换原始内容 */
+    private static final String SAFE_AI_RESPONSE = "抱歉，该内容无法展示。如有问题请联系管理员。";
+
+    @Autowired
+    private ContentSafetyService contentSafetyService;
+
+    @Autowired
+    private ContentAuditLogMapper contentAuditLogMapper;
+
+    @Override
+    public String getName() {
+        return this.getClass().getSimpleName();
+    }
+
+    @Override
+    public int getOrder() {
+        return Ordered.HIGHEST_PRECEDENCE;
+    }
+
+    // ==================== before：检查用户输入 ====================
+
+    @Override
+    public ChatClientRequest before(ChatClientRequest request, AdvisorChain chain) {
+        try {
+            Prompt originalPrompt = request.prompt();
+            List<Message> messages = new ArrayList<>(originalPrompt.getInstructions());
+
+            // 提取最后一条用户消息进行检测
+            String userText = null;
+            int lastUserIndex = -1;
+            for (int i = messages.size() - 1; i >= 0; i--) {
+                if (messages.get(i) instanceof UserMessage) {
+                    userText = messages.get(i).getText();
+                    lastUserIndex = i;
+                    break;
+                }
+            }
+
+            if (userText == null || userText.isBlank()) {
+                return request;
+            }
+
+            // 执行 DFA 检测
+            List<ContentSafetyService.HitResult> hits = contentSafetyService.detect(userText);
+            if (hits.isEmpty()) {
+                return request;
+            }
+
+            boolean hasBlocking = contentSafetyService.hasBlockingHit(hits);
+            String sessionId = extractSessionId(request);
+
+            if (hasBlocking) {
+                // BLOCK 级别：替换用户消息为安全提示，记录审计日志
+                log.warn("内容安全拦截（BLOCK）- 会话：{}，命中词：{}", sessionId,
+                        hits.stream().map(ContentSafetyService.HitResult::getWord).collect(Collectors.joining(", ")));
+
+                messages.set(lastUserIndex, new UserMessage(SAFE_USER_PROMPT));
+                Prompt newPrompt = new Prompt(messages, originalPrompt.getOptions());
+
+                saveAuditLog(sessionId, "INPUT", truncate(userText), hits, "BLOCK");
+
+                return ChatClientRequest.builder()
+                        .prompt(newPrompt)
+                        .context(request.context())
+                        .build();
+            } else {
+                // WARN 级别：脱敏后替换用户消息，记录审计日志
+                String maskedText = contentSafetyService.mask(userText);
+                log.info("内容安全脱敏（MASK）- 会话：{}，命中词：{}", sessionId,
+                        hits.stream().map(ContentSafetyService.HitResult::getWord).collect(Collectors.joining(", ")));
+
+                messages.set(lastUserIndex, new UserMessage(maskedText));
+                Prompt newPrompt = new Prompt(messages, originalPrompt.getOptions());
+
+                saveAuditLog(sessionId, "INPUT", truncate(userText), hits, "MASK");
+
+                return ChatClientRequest.builder()
+                        .prompt(newPrompt)
+                        .context(request.context())
+                        .build();
+            }
+        } catch (Exception e) {
+            log.error("内容安全 before 检查异常，放行请求", e);
+            return request;
+        }
+    }
+
+    // ==================== after：检查 AI 输出 ====================
+
+    @Override
+    public ChatClientResponse after(ChatClientResponse response, AdvisorChain chain) {
+        try {
+            ChatResponse chatResponse = response.chatResponse();
+            if (chatResponse == null || chatResponse.getResult() == null) {
+                return response;
+            }
+
+            String aiText = chatResponse.getResult().getOutput().getText();
+            if (aiText == null || aiText.isBlank()) {
+                return response;
+            }
+
+            // 执行 DFA 检测
+            List<ContentSafetyService.HitResult> hits = contentSafetyService.detect(aiText);
+            if (hits.isEmpty()) {
+                return response;
+            }
+
+            boolean hasBlocking = contentSafetyService.hasBlockingHit(hits);
+            String actionTaken = hasBlocking ? "BLOCK" : "MASK";
+
+            log.warn("AI 输出内容安全拦截（{}）- 命中词：{}", actionTaken,
+                    hits.stream().map(ContentSafetyService.HitResult::getWord).collect(Collectors.joining(", ")));
+
+            // 构建替换后的 AI 响应
+            String safeText = hasBlocking ? SAFE_AI_RESPONSE : contentSafetyService.mask(aiText);
+            AssistantMessage safeAssistantMessage = new AssistantMessage(safeText);
+            Generation safeGeneration = new Generation(safeAssistantMessage, chatResponse.getResult().getMetadata());
+            ChatResponse safeChatResponse = new ChatResponse(
+                    Collections.singletonList(safeGeneration),
+                    chatResponse.getMetadata()
+            );
+
+            // 保存审计日志（此处 sessionId 可能为空，因为 after 阶段不一定能获取到）
+            saveAuditLog("", "OUTPUT", truncate(aiText), hits, actionTaken);
+
+            return ChatClientResponse.builder()
+                    .chatResponse(safeChatResponse)
+                    .context(response.context())
+                    .build();
+        } catch (Exception e) {
+            log.error("内容安全 after 检查异常，放行响应", e);
+            return response;
+        }
+    }
+
+    // ==================== 辅助方法 ====================
+
+    /**
+     * 从请求上下文中提取会话ID
+     */
+    private String extractSessionId(ChatClientRequest request) {
+        Map<String, Object> context = request.context();
+        if (context != null && context.containsKey("chatMemoryConversationId")) {
+            return String.valueOf(context.get("chatMemoryConversationId"));
+        }
+        return "";
+    }
+
+    /**
+     * 截断文本至指定长度
+     */
+    private String truncate(String text) {
+        if (text == null) {
+            return "";
+        }
+        return text.length() > 50 ? text.substring(0, 50) + "..." : text;
+    }
+
+    /**
+     * 保存审计日志
+     */
+    private void saveAuditLog(String sessionId, String direction, String originalText,
+                              List<ContentSafetyService.HitResult> hits, String actionTaken) {
+        try {
+            // 将命中结果转为 Map 列表存入 JSONB
+            List<Map<String, Object>> hitWordsList = hits.stream().map(hit -> {
+                Map<String, Object> m = new LinkedHashMap<>();
+                m.put("word", hit.getWord());
+                m.put("category", hit.getCategory());
+                m.put("level", hit.getLevel());
+                return m;
+            }).collect(Collectors.toList());
+
+            Map<String, Object> hitWordsMap = new LinkedHashMap<>();
+            hitWordsMap.put("hits", hitWordsList);
+
+            ContentAuditLog auditLog = ContentAuditLog.builder()
+                    .sessionId(sessionId)
+                    .direction(direction)
+                    .originalText(originalText)
+                    .hitWords(hitWordsMap)
+                    .actionTaken(actionTaken)
+                    .createTime(new Date())
+                    .build();
+
+            contentAuditLogMapper.insert(auditLog);
+        } catch (Exception e) {
+            log.error("保存审计日志失败", e);
+        }
+    }
+}

src/main/java/com/wok/supportbot/app/AssistantApp.java

@@ -1,5 +1,6 @@
 package com.wok.supportbot.app;
 
+import com.wok.supportbot.advisor.ContentSafetyAdvisor;
 import com.wok.supportbot.advisor.MyLoggerAdvisor;
 import com.wok.supportbot.advisor.ReReadingAdvisor;
 import com.wok.supportbot.chatmemory.DatabaseChatMemory;
@@ -56,6 +57,9 @@ public class AssistantApp {
     @Resource
     private VectorStore pgVectorVectorStore;
 
+    @Resource
+    private ContentSafetyAdvisor contentSafetyAdvisor;
+
     private final ChatModelFactory chatModelFactory;
 
     private final DatabaseChatMemory chatMemory;
@@ -81,6 +85,7 @@ public class AssistantApp {
             return ChatClient.builder(chatModel)
                     .defaultSystem(SYSTEM_PROMPT)
                     .defaultAdvisors(
+                            contentSafetyAdvisor,
                             MessageChatMemoryAdvisor.builder(chatMemory).build(),
                             new MyLoggerAdvisor()
                     )

src/main/java/com/wok/supportbot/config/DatabaseInitConfig.java

@@ -3,6 +3,7 @@ package com.wok.supportbot.config;
 import jakarta.annotation.PostConstruct;
 import lombok.extern.slf4j.Slf4j;
 import org.springframework.beans.factory.annotation.Autowired;
+import org.springframework.beans.factory.annotation.Value;
 import org.springframework.jdbc.core.JdbcTemplate;
 import org.springframework.stereotype.Component;
 
@@ -17,6 +18,9 @@ public class DatabaseInitConfig {
     @Autowired
     private JdbcTemplate jdbcTemplate;
 
+    @Value("${knowledge.vector.dimension:1024}")
+    private int vectorDimension;
+
     @PostConstruct
     public void init() {
         try {
@@ -75,6 +79,40 @@ public class DatabaseInitConfig {
             syncDefaultCustomerServiceRoles();
             syncDefaultCustomerAccounts();
 
+            // ==================== P0 阶段新增表 ====================
+
+            // P0-004: 内容安全过滤
+            if (!checkTableExists("sensitive_word")) {
+                log.info("创建敏感词表 sensitive_word");
+                createSensitiveWordTable();
+            }
+            if (!checkTableExists("content_audit_log")) {
+                log.info("创建内容审计日志表 content_audit_log");
+                createContentAuditLogTable();
+            }
+
+            // P0-002: 用户反馈
+            if (!checkTableExists("message_feedback")) {
+                log.info("创建消息反馈表 message_feedback");
+                createMessageFeedbackTable();
+            }
+
+            // P0-003: FAQ 知识库
+            if (!checkTableExists("knowledge_faq")) {
+                log.info("创建 FAQ 知识库表 knowledge_faq");
+                createKnowledgeFaqTable();
+            }
+            if (!checkTableExists("faq_embedding")) {
+                log.info("创建 FAQ 向量索引表 faq_embedding");
+                createFaqEmbeddingTable();
+            }
+
+            // P0-001: 混合检索 - 为 vector_store 添加全文检索列
+            initVectorStoreFullTextSearch();
+
+            // 为所有表添加注释（幂等，可重复执行）
+            applyTableComments();
+
             log.info("数据库初始化完成");
         } catch (Exception e) {
             log.error("数据库初始化失败", e);
@@ -348,4 +386,317 @@ public class DatabaseInitConfig {
         }
     }
 
+    // ==================== P0-004: 内容安全过滤 ====================
+
+    private void createSensitiveWordTable() {
+        String sql = """
+            CREATE TABLE IF NOT EXISTS sensitive_word (
+                id              BIGSERIAL       PRIMARY KEY,
+                word            VARCHAR(256)    NOT NULL,
+                category        VARCHAR(64)     NOT NULL DEFAULT 'custom',
+                level           INTEGER         NOT NULL DEFAULT 1,
+                is_active       BOOLEAN         NOT NULL DEFAULT TRUE,
+                remark          VARCHAR(512),
+                create_time     TIMESTAMP       DEFAULT CURRENT_TIMESTAMP NOT NULL,
+                update_time     TIMESTAMP       DEFAULT CURRENT_TIMESTAMP NOT NULL,
+                is_delete       BOOLEAN         DEFAULT FALSE             NOT NULL
+            )
+            """;
+        jdbcTemplate.execute(sql);
+        jdbcTemplate.execute("CREATE UNIQUE INDEX IF NOT EXISTS uk_sensitive_word_word_category ON sensitive_word (word, category) WHERE is_delete = false");
+        jdbcTemplate.execute("CREATE INDEX IF NOT EXISTS idx_sensitive_word_active ON sensitive_word (is_active) WHERE is_delete = false");
+    }
+
+    private void createContentAuditLogTable() {
+        String sql = """
+            CREATE TABLE IF NOT EXISTS content_audit_log (
+                id              BIGSERIAL       PRIMARY KEY,
+                session_id      VARCHAR(64),
+                direction       VARCHAR(16)     NOT NULL,
+                original_text   TEXT            NOT NULL,
+                hit_words       JSONB           DEFAULT '[]' NOT NULL,
+                action_taken    VARCHAR(32)     NOT NULL,
+                create_time     TIMESTAMP       DEFAULT CURRENT_TIMESTAMP NOT NULL
+            )
+            """;
+        jdbcTemplate.execute(sql);
+        jdbcTemplate.execute("CREATE INDEX IF NOT EXISTS idx_audit_log_created ON content_audit_log (create_time DESC)");
+        jdbcTemplate.execute("CREATE INDEX IF NOT EXISTS idx_audit_log_session ON content_audit_log (session_id)");
+    }
+
+    // ==================== P0-002: 用户反馈 ====================
+
+    private void createMessageFeedbackTable() {
+        String sql = """
+            CREATE TABLE IF NOT EXISTS message_feedback (
+                id              BIGSERIAL       PRIMARY KEY,
+                message_id      VARCHAR(128)    NOT NULL,
+                conversation_id VARCHAR(64)     NOT NULL,
+                feedback_type   VARCHAR(16)     NOT NULL,
+                reason_category VARCHAR(64),
+                reason_comment  TEXT,
+                create_time     TIMESTAMP       DEFAULT CURRENT_TIMESTAMP NOT NULL,
+                update_time     TIMESTAMP       DEFAULT CURRENT_TIMESTAMP NOT NULL,
+                is_delete       BOOLEAN         DEFAULT FALSE             NOT NULL
+            )
+            """;
+        jdbcTemplate.execute(sql);
+        jdbcTemplate.execute("CREATE UNIQUE INDEX IF NOT EXISTS uk_message_feedback_message ON message_feedback (message_id) WHERE is_delete = false");
+        jdbcTemplate.execute("CREATE INDEX IF NOT EXISTS idx_feedback_conversation ON message_feedback (conversation_id)");
+        jdbcTemplate.execute("CREATE INDEX IF NOT EXISTS idx_feedback_type ON message_feedback (feedback_type) WHERE is_delete = false");
+        jdbcTemplate.execute("CREATE INDEX IF NOT EXISTS idx_feedback_created ON message_feedback (create_time DESC)");
+    }
+
+    // ==================== P0-003: FAQ 知识库 ====================
+
+    private void createKnowledgeFaqTable() {
+        String sql = """
+            CREATE TABLE IF NOT EXISTS knowledge_faq (
+                id                  BIGSERIAL       PRIMARY KEY,
+                question            TEXT            NOT NULL,
+                answer              TEXT            NOT NULL,
+                similar_questions   TEXT            DEFAULT '[]' NOT NULL,
+                category            VARCHAR(128),
+                status              VARCHAR(20)     NOT NULL DEFAULT 'ENABLED',
+                priority            INTEGER         NOT NULL DEFAULT 0,
+                hit_count           BIGINT          NOT NULL DEFAULT 0,
+                source              VARCHAR(64)     DEFAULT 'manual',
+                create_time         TIMESTAMP       DEFAULT CURRENT_TIMESTAMP NOT NULL,
+                update_time         TIMESTAMP       DEFAULT CURRENT_TIMESTAMP NOT NULL,
+                is_delete           BOOLEAN         DEFAULT FALSE             NOT NULL
+            )
+            """;
+        jdbcTemplate.execute(sql);
+        jdbcTemplate.execute("CREATE INDEX IF NOT EXISTS idx_faq_status ON knowledge_faq (status) WHERE is_delete = false");
+        jdbcTemplate.execute("CREATE INDEX IF NOT EXISTS idx_faq_category ON knowledge_faq (category)");
+        jdbcTemplate.execute("CREATE INDEX IF NOT EXISTS idx_faq_priority ON knowledge_faq (priority DESC)");
+    }
+
+    private void createFaqEmbeddingTable() {
+        String sql = """
+            CREATE TABLE IF NOT EXISTS faq_embedding (
+                id              BIGSERIAL       PRIMARY KEY,
+                faq_id          BIGINT          NOT NULL,
+                embedding       vector(%d)      NOT NULL,
+                model_name      VARCHAR(64)     NOT NULL,
+                create_time     TIMESTAMP       DEFAULT CURRENT_TIMESTAMP NOT NULL
+            )
+            """.formatted(vectorDimension);
+        jdbcTemplate.execute(sql);
+        jdbcTemplate.execute("CREATE INDEX IF NOT EXISTS idx_faq_emb_faq_id ON faq_embedding (faq_id)");
+    }
+
+    // ==================== P0-001: 混合检索 - 全文检索 ====================
+
+    /**
+     * 为 vector_store 表添加全文检索支持（tsvector 列 + GIN 索引 + 触发器）
+     * 幂等操作：已存在则跳过
+     */
+    private void initVectorStoreFullTextSearch() {
+        try {
+            // 检查 vector_store 表是否存在
+            if (!checkTableExists("vector_store")) {
+                log.debug("vector_store 表尚未创建，跳过全文检索初始化");
+                return;
+            }
+
+            // 检查 content_tsvector 列是否已存在
+            String checkSql = "SELECT COUNT(*) FROM information_schema.columns WHERE table_name = 'vector_store' AND column_name = 'content_tsvector'";
+            Integer count = jdbcTemplate.queryForObject(checkSql, Integer.class);
+            if (count != null && count > 0) {
+                return; // 已初始化
+            }
+
+            log.info("为 vector_store 添加全文检索支持");
+
+            // 添加 tsvector 列
+            jdbcTemplate.execute("ALTER TABLE vector_store ADD COLUMN content_tsvector tsvector");
+
+            // 填充已有数据
+            jdbcTemplate.execute("UPDATE vector_store SET content_tsvector = to_tsvector('simple', coalesce(content, '')) WHERE content_tsvector IS NULL");
+
+            // GIN 索引
+            jdbcTemplate.execute("CREATE INDEX IF NOT EXISTS idx_vector_store_tsvector ON vector_store USING gin(content_tsvector)");
+
+            // 触发器函数
+            jdbcTemplate.execute("""
+                CREATE OR REPLACE FUNCTION update_content_tsvector() RETURNS trigger AS $$
+                BEGIN
+                    NEW.content_tsvector := to_tsvector('simple', coalesce(NEW.content, ''));
+                    RETURN NEW;
+                END;
+                $$ LANGUAGE plpgsql
+                """);
+
+            // 绑定触发器
+            jdbcTemplate.execute("DROP TRIGGER IF EXISTS trg_update_content_tsvector ON vector_store");
+            jdbcTemplate.execute("""
+                CREATE TRIGGER trg_update_content_tsvector
+                    BEFORE INSERT OR UPDATE ON vector_store
+                    FOR EACH ROW EXECUTE FUNCTION update_content_tsvector()
+                """);
+
+            log.info("vector_store 全文检索初始化完成");
+        } catch (Exception e) {
+            log.warn("初始化 vector_store 全文检索时出错（可稍后手动执行）", e);
+        }
+    }
+
+    /**
+     * 为所有自动创建的表添加注释（COMMENT ON）。
+     * 所有语句均为幂等操作，可安全重复执行。
+     */
+    private void applyTableComments() {
+        try {
+            // ===== chat_message =====
+            executeComment("TABLE chat_message", "聊天消息表（存储用户与 AI 助手的对话历史）");
+            executeComment("COLUMN chat_message.id", "主键（雪花算法生成）");
+            executeComment("COLUMN chat_message.conversation_id", "会话 ID（标识同一次对话）");
+            executeComment("COLUMN chat_message.message_type", "消息类型: USER(用户消息) / ASSISTANT(AI回复) / SYSTEM(系统消息)");
+            executeComment("COLUMN chat_message.content", "消息内容（实际对话文本）");
+            executeComment("COLUMN chat_message.metadata", "元数据（JSON 格式，存储额外信息）");
+            executeComment("COLUMN chat_message.create_time", "创建时间");
+            executeComment("COLUMN chat_message.update_time", "更新时间");
+            executeComment("COLUMN chat_message.is_delete", "逻辑删除: FALSE=正常 TRUE=已删除");
+
+            // ===== knowledge_category =====
+            executeComment("TABLE knowledge_category", "知识库分类表（支持树形结构）");
+            executeComment("COLUMN knowledge_category.id", "主键（雪花算法生成）");
+            executeComment("COLUMN knowledge_category.name", "分类名称");
+            executeComment("COLUMN knowledge_category.description", "分类描述");
+            executeComment("COLUMN knowledge_category.parent_id", "父分类 ID（0 表示顶级分类）");
+            executeComment("COLUMN knowledge_category.sort_order", "排序权重（数值越大越靠前）");
+            executeComment("COLUMN knowledge_category.document_count", "关联文档数量（冗余字段，定期更新）");
+            executeComment("COLUMN knowledge_category.create_time", "创建时间");
+            executeComment("COLUMN knowledge_category.is_delete", "逻辑删除: FALSE=正常 TRUE=已删除");
+
+            // ===== knowledge_document =====
+            executeComment("TABLE knowledge_document", "知识文档表（记录上传的文档元信息）");
+            executeComment("COLUMN knowledge_document.id", "主键（雪花算法生成）");
+            executeComment("COLUMN knowledge_document.title", "文档标题");
+            executeComment("COLUMN knowledge_document.source_name", "原始文件名");
+            executeComment("COLUMN knowledge_document.file_type", "文件类型: pdf / md / json / txt / word / excel 等");
+            executeComment("COLUMN knowledge_document.file_size", "文件大小（字节）");
+            executeComment("COLUMN knowledge_document.content", "原文内容（截断预览）");
+            executeComment("COLUMN knowledge_document.category_id", "所属分类 ID（0 表示未分类）");
+            executeComment("COLUMN knowledge_document.tags", "标签（JSON 格式）");
+            executeComment("COLUMN knowledge_document.chunk_count", "分块数量");
+            executeComment("COLUMN knowledge_document.status", "处理状态: PROCESSING / READY / FAILED");
+            executeComment("COLUMN knowledge_document.error_message", "处理失败时的错误信息");
+            executeComment("COLUMN knowledge_document.content_hash", "内容 SHA-256 哈希值（用于文档去重）");
+            executeComment("COLUMN knowledge_document.create_time", "创建时间");
+            executeComment("COLUMN knowledge_document.update_time", "更新时间");
+            executeComment("COLUMN knowledge_document.is_delete", "逻辑删除: FALSE=正常 TRUE=已删除");
+
+            // ===== customer_service_role =====
+            executeComment("TABLE customer_service_role", "客服角色表（定义客服角色的身份与系统提示词）");
+            executeComment("COLUMN customer_service_role.id", "主键");
+            executeComment("COLUMN customer_service_role.role_key", "角色标识符（唯一）");
+            executeComment("COLUMN customer_service_role.name", "角色名称");
+            executeComment("COLUMN customer_service_role.description", "角色描述");
+            executeComment("COLUMN customer_service_role.prompt", "系统提示词（角色人设与行为规范）");
+            executeComment("COLUMN customer_service_role.sort_order", "排序权重");
+            executeComment("COLUMN customer_service_role.enabled", "是否启用");
+            executeComment("COLUMN customer_service_role.create_time", "创建时间");
+            executeComment("COLUMN customer_service_role.update_time", "更新时间");
+            executeComment("COLUMN customer_service_role.is_delete", "逻辑删除: FALSE=正常 TRUE=已删除");
+
+            // ===== customer_service_role_category =====
+            executeComment("TABLE customer_service_role_category", "客服角色知识库关联表（角色与知识库分类的多对多关系）");
+            executeComment("COLUMN customer_service_role_category.id", "主键");
+            executeComment("COLUMN customer_service_role_category.role_id", "角色 ID（关联 customer_service_role.id）");
+            executeComment("COLUMN customer_service_role_category.category_id", "分类 ID（关联 knowledge_category.id）");
+            executeComment("COLUMN customer_service_role_category.create_time", "创建时间");
+            executeComment("COLUMN customer_service_role_category.is_delete", "逻辑删除: FALSE=正常 TRUE=已删除");
+
+            // ===== customer_account =====
+            executeComment("TABLE customer_account", "客服账号表（对外暴露的客服入口账号）");
+            executeComment("COLUMN customer_account.id", "主键");
+            executeComment("COLUMN customer_account.account_key", "账号标识符（唯一）");
+            executeComment("COLUMN customer_account.name", "账号名称");
+            executeComment("COLUMN customer_account.description", "账号描述");
+            executeComment("COLUMN customer_account.role_id", "关联角色 ID（关联 customer_service_role.id）");
+            executeComment("COLUMN customer_account.enabled", "是否启用");
+            executeComment("COLUMN customer_account.create_time", "创建时间");
+            executeComment("COLUMN customer_account.update_time", "更新时间");
+            executeComment("COLUMN customer_account.is_delete", "逻辑删除: FALSE=正常 TRUE=已删除");
+
+            // ===== conversation_session =====
+            executeComment("TABLE conversation_session", "会话归属表（记录每个会话归属的账号和角色）");
+            executeComment("COLUMN conversation_session.conversation_id", "会话 ID（主键）");
+            executeComment("COLUMN conversation_session.account_id", "归属账号 ID（关联 customer_account.id）");
+            executeComment("COLUMN conversation_session.role_id", "归属角色 ID（关联 customer_service_role.id）");
+            executeComment("COLUMN conversation_session.create_time", "创建时间");
+            executeComment("COLUMN conversation_session.update_time", "更新时间");
+
+            // ===== sensitive_word =====
+            executeComment("TABLE sensitive_word", "敏感词表（DFA 引擎驱动的内容安全过滤）");
+            executeComment("COLUMN sensitive_word.id", "主键（雪花算法生成）");
+            executeComment("COLUMN sensitive_word.word", "敏感词内容");
+            executeComment("COLUMN sensitive_word.category", "分类: politics / porn / abuse / custom，默认 custom");
+            executeComment("COLUMN sensitive_word.level", "级别: 1=脱敏(MASK) / 2=拦截(BLOCK)");
+            executeComment("COLUMN sensitive_word.is_active", "是否启用");
+            executeComment("COLUMN sensitive_word.remark", "备注说明");
+            executeComment("COLUMN sensitive_word.create_time", "创建时间");
+            executeComment("COLUMN sensitive_word.update_time", "更新时间");
+            executeComment("COLUMN sensitive_word.is_delete", "逻辑删除: FALSE=正常 TRUE=已删除");
+
+            // ===== content_audit_log =====
+            executeComment("TABLE content_audit_log", "内容审计日志表（记录敏感词命中事件，只追加不删除）");
+            executeComment("COLUMN content_audit_log.id", "主键（雪花算法生成）");
+            executeComment("COLUMN content_audit_log.session_id", "会话 ID（关联 conversation_session）");
+            executeComment("COLUMN content_audit_log.direction", "检测方向: INPUT(用户输入) / OUTPUT(AI 输出)");
+            executeComment("COLUMN content_audit_log.original_text", "原始违规内容（截断至前 50 字）");
+            executeComment("COLUMN content_audit_log.hit_words", "命中词列表（JSON 数组）");
+            executeComment("COLUMN content_audit_log.action_taken", "采取的动作: PASS(放行) / MASK(脱敏) / BLOCK(拦截)");
+            executeComment("COLUMN content_audit_log.create_time", "创建时间（事件发生时间）");
+
+            // ===== message_feedback =====
+            executeComment("TABLE message_feedback", "消息反馈表（用户对 AI 回复的点赞/点踩反馈）");
+            executeComment("COLUMN message_feedback.id", "主键（雪花算法生成）");
+            executeComment("COLUMN message_feedback.message_id", "AI 消息 ID（唯一索引，重复提交覆盖）");
+            executeComment("COLUMN message_feedback.conversation_id", "会话 ID");
+            executeComment("COLUMN message_feedback.feedback_type", "反馈类型: THUMBS_UP(有帮助) / THUMBS_DOWN(没帮助)");
+            executeComment("COLUMN message_feedback.reason_category", "点踩原因分类: inaccurate / irrelevant / incomplete / other");
+            executeComment("COLUMN message_feedback.reason_comment", "自由文本补充说明");
+            executeComment("COLUMN message_feedback.create_time", "创建时间");
+            executeComment("COLUMN message_feedback.update_time", "更新时间（重复提交时覆盖）");
+            executeComment("COLUMN message_feedback.is_delete", "逻辑删除: FALSE=正常 TRUE=已删除");
+
+            // ===== knowledge_faq =====
+            executeComment("TABLE knowledge_faq", "FAQ 知识库表（用于意图路由后的精准问答匹配）");
+            executeComment("COLUMN knowledge_faq.id", "主键（雪花算法生成）");
+            executeComment("COLUMN knowledge_faq.question", "标准问题");
+            executeComment("COLUMN knowledge_faq.answer", "标准答案");
+            executeComment("COLUMN knowledge_faq.similar_questions", "相似问题列表（JSON 字符串数组）");
+            executeComment("COLUMN knowledge_faq.category", "FAQ 分类");
+            executeComment("COLUMN knowledge_faq.status", "状态: ENABLED(启用) / DISABLED(禁用)");
+            executeComment("COLUMN knowledge_faq.priority", "优先级（数值越大越优先匹配）");
+            executeComment("COLUMN knowledge_faq.hit_count", "命中次数统计");
+            executeComment("COLUMN knowledge_faq.source", "来源: manual(手动录入) / import(批量导入)");
+            executeComment("COLUMN knowledge_faq.create_time", "创建时间");
+            executeComment("COLUMN knowledge_faq.update_time", "更新时间");
+            executeComment("COLUMN knowledge_faq.is_delete", "逻辑删除: FALSE=正常 TRUE=已删除");
+
+            // ===== faq_embedding =====
+            executeComment("TABLE faq_embedding", "FAQ 向量索引表（存储 FAQ 语义向量，用于相似问题匹配）");
+            executeComment("COLUMN faq_embedding.id", "主键");
+            executeComment("COLUMN faq_embedding.faq_id", "关联 FAQ ID（对应 knowledge_faq.id）");
+            executeComment("COLUMN faq_embedding.embedding", "向量嵌入（维度由 knowledge.vector.dimension 配置）");
+            executeComment("COLUMN faq_embedding.model_name", "使用的 Embedding 模型名称");
+            executeComment("COLUMN faq_embedding.create_time", "创建时间");
+
+            log.info("数据库表注释已应用");
+        } catch (Exception e) {
+            log.warn("应用数据库表注释时出错", e);
+        }
+    }
+
+    /**
+     * 执行单条 COMMENT ON 语句
+     */
+    private void executeComment(String target, String comment) {
+        jdbcTemplate.execute(String.format("COMMENT ON %s IS '%s'", target, comment.replace("'", "''")));
+    }
+
 }

src/main/java/com/wok/supportbot/controller/AiController.java

@@ -8,6 +8,7 @@ import com.wok.supportbot.service.CustomerAccountService.AccountScope;
 import com.wok.supportbot.service.CustomerServiceRoleService;
 import com.wok.supportbot.service.CustomerServiceRoleService.RoleScope;
 import jakarta.annotation.Resource;
+import lombok.extern.slf4j.Slf4j;
 import org.springframework.ai.document.Document;
 import org.springframework.http.MediaType;
 import org.springframework.http.codec.ServerSentEvent;
@@ -29,6 +30,7 @@ import java.util.Map;
 
 
 @RestController
+@Slf4j
 @RequestMapping("/ai")
 public class AiController {
 
@@ -146,9 +148,14 @@ public class AiController {
         if (isKbDenied(scope) || shouldBypassKnowledgeRetrieval(message)) {
             return assistantApp.doChat(message, chatId, sys);
         }
-        return assistantApp.doChatWithRagStrategy(
-                message, chatId, normalizeStrategy(rewriteStrategy),
-                resolveCategoryIds(scope, categoryId, categoryIds), sys);
+        try {
+            return assistantApp.doChatWithRagStrategy(
+                    message, chatId, normalizeStrategy(rewriteStrategy),
+                    resolveCategoryIds(scope, categoryId, categoryIds), sys);
+        } catch (Exception e) {
+            log.error("RAG 对话失败 [strategy={}, chatId={}]: {}", rewriteStrategy, chatId, e.getMessage(), e);
+            return "抱歉，知识库检索出现异常，请稍后重试。";
+        }
     }
 
     /**
@@ -190,22 +197,27 @@ public class AiController {
         if (message == null || message.isBlank() || isKbDenied(scope) || shouldBypassKnowledgeRetrieval(message)) {
             return Map.of("success", true, "data", List.of());
         }
-        List<Long> cats = resolveCategoryIds(scope, categoryId, categoryIds);
-        List<Document> docs = assistantApp.retrieveRagSources(message, chatId, normalizeStrategy(rewriteStrategy), cats);
-        List<Map<String, Object>> out = new ArrayList<>();
-        for (Document doc : docs) {
-            Map<String, Object> meta = doc.getMetadata();
-            Map<String, Object> item = new LinkedHashMap<>();
-            item.put("documentId", meta.get("documentId"));
-            item.put("title", meta.get("title"));
-            item.put("sourceName", meta.get("sourceName"));
-            item.put("chunkIndex", meta.get("chunkIndex"));
-            item.put("score", meta.get("distance"));
-            String text = doc.getText();
-            item.put("snippet", text != null && text.length() > 160 ? text.substring(0, 160) + "…" : text);
-            out.add(item);
+        try {
+            List<Long> cats = resolveCategoryIds(scope, categoryId, categoryIds);
+            List<Document> docs = assistantApp.retrieveRagSources(message, chatId, normalizeStrategy(rewriteStrategy), cats);
+            List<Map<String, Object>> out = new ArrayList<>();
+            for (Document doc : docs) {
+                Map<String, Object> meta = doc.getMetadata();
+                Map<String, Object> item = new LinkedHashMap<>();
+                item.put("documentId", meta.get("documentId"));
+                item.put("title", meta.get("title"));
+                item.put("sourceName", meta.get("sourceName"));
+                item.put("chunkIndex", meta.get("chunkIndex"));
+                item.put("score", meta.get("distance"));
+                String text = doc.getText();
+                item.put("snippet", text != null && text.length() > 160 ? text.substring(0, 160) + "…" : text);
+                out.add(item);
+            }
+            return Map.of("success", true, "data", out);
+        } catch (Exception e) {
+            log.error("获取 RAG 引用来源失败 [strategy={}]: {}", rewriteStrategy, e.getMessage(), e);
+            return Map.of("success", true, "data", List.of());
         }
-        return Map.of("success", true, "data", out);
     }
 
     /**

src/main/java/com/wok/supportbot/controller/DocumentController.java

@@ -468,8 +468,14 @@ public class DocumentController {
             if (categoryIds.isEmpty() && categoryId != null) {
                 categoryIds = List.of(categoryId);
             }
+            String searchMode = (String) body.get("searchMode");
 
-            List<SearchResult> results = documentService.searchDocuments(query, topK, similarityThreshold, categoryIds);
+            List<SearchResult> results;
+            if (searchMode != null && !searchMode.isEmpty()) {
+                results = documentService.searchDocuments(query, topK, similarityThreshold, categoryIds, searchMode);
+            } else {
+                results = documentService.searchDocuments(query, topK, similarityThreshold, categoryIds);
+            }
             return ResponseEntity.ok(Map.of(
                     "success", true,
                     "data", results,

src/main/java/com/wok/supportbot/controller/FaqController.java

@@ -0,0 +1,277 @@
+package com.wok.supportbot.controller;
+
+import com.wok.supportbot.entity.KnowledgeFaq;
+import com.wok.supportbot.service.FaqService;
+import lombok.extern.slf4j.Slf4j;
+import org.springframework.beans.factory.annotation.Autowired;
+import org.springframework.http.ResponseEntity;
+import org.springframework.web.bind.annotation.*;
+
+import java.util.List;
+import java.util.Map;
+
+/**
+ * FAQ 知识库管理控制器
+ * 提供 FAQ 的 CRUD、批量导入/导出、匹配统计等功能
+ */
+@RestController
+@RequestMapping("/faq")
+@Slf4j
+public class FaqController {
+
+    @Autowired
+    private FaqService faqService;
+
+    // ==================== 分页列表 ====================
+
+    /**
+     * 分页查询 FAQ 列表
+     */
+    @GetMapping("/list")
+    public ResponseEntity<Map<String, Object>> list(
+            @RequestParam(defaultValue = "1") int page,
+            @RequestParam(defaultValue = "20") int size,
+            @RequestParam(required = false) String keyword,
+            @RequestParam(required = false) String category,
+            @RequestParam(required = false) String status) {
+        try {
+            Map<String, Object> data = faqService.list(page, size, keyword, category, status);
+            return ResponseEntity.ok(Map.of(
+                    "success", true,
+                    "message", "查询成功",
+                    "data", data
+            ));
+        } catch (Exception e) {
+            log.error("FAQ 列表查询失败", e);
+            return ResponseEntity.status(500).body(Map.of(
+                    "success", false,
+                    "message", "查询失败：" + e.getMessage()
+            ));
+        }
+    }
+
+    // ==================== 新增 ====================
+
+    /**
+     * 新增 FAQ
+     */
+    @PostMapping
+    public ResponseEntity<Map<String, Object>> create(@RequestBody KnowledgeFaq faq) {
+        try {
+            if (faq.getQuestion() == null || faq.getQuestion().isBlank()) {
+                return ResponseEntity.status(400).body(Map.of(
+                        "success", false,
+                        "message", "问题内容不能为空"
+                ));
+            }
+            if (faq.getAnswer() == null || faq.getAnswer().isBlank()) {
+                return ResponseEntity.status(400).body(Map.of(
+                        "success", false,
+                        "message", "答案内容不能为空"
+                ));
+            }
+
+            KnowledgeFaq created = faqService.create(faq);
+            return ResponseEntity.ok(Map.of(
+                    "success", true,
+                    "message", "创建成功",
+                    "data", created
+            ));
+        } catch (Exception e) {
+            log.error("FAQ 创建失败", e);
+            return ResponseEntity.status(500).body(Map.of(
+                    "success", false,
+                    "message", "创建失败：" + e.getMessage()
+            ));
+        }
+    }
+
+    // ==================== 修改 ====================
+
+    /**
+     * 修改 FAQ
+     */
+    @PutMapping("/{id}")
+    public ResponseEntity<Map<String, Object>> update(@PathVariable Long id, @RequestBody KnowledgeFaq faq) {
+        try {
+            KnowledgeFaq updated = faqService.update(id, faq);
+            return ResponseEntity.ok(Map.of(
+                    "success", true,
+                    "message", "更新成功",
+                    "data", updated
+            ));
+        } catch (IllegalArgumentException e) {
+            return ResponseEntity.status(400).body(Map.of(
+                    "success", false,
+                    "message", e.getMessage()
+            ));
+        } catch (Exception e) {
+            log.error("FAQ 更新失败: id={}", id, e);
+            return ResponseEntity.status(500).body(Map.of(
+                    "success", false,
+                    "message", "更新失败：" + e.getMessage()
+            ));
+        }
+    }
+
+    // ==================== 删除 ====================
+
+    /**
+     * 删除 FAQ（逻辑删除）
+     */
+    @DeleteMapping("/{id}")
+    public ResponseEntity<Map<String, Object>> delete(@PathVariable Long id) {
+        try {
+            faqService.delete(id);
+            return ResponseEntity.ok(Map.of(
+                    "success", true,
+                    "message", "删除成功"
+            ));
+        } catch (IllegalArgumentException e) {
+            return ResponseEntity.status(400).body(Map.of(
+                    "success", false,
+                    "message", e.getMessage()
+            ));
+        } catch (Exception e) {
+            log.error("FAQ 删除失败: id={}", id, e);
+            return ResponseEntity.status(500).body(Map.of(
+                    "success", false,
+                    "message", "删除失败：" + e.getMessage()
+            ));
+        }
+    }
+
+    // ==================== 启用/禁用 ====================
+
+    /**
+     * 切换 FAQ 启用/禁用状态
+     */
+    @PutMapping("/{id}/toggle")
+    public ResponseEntity<Map<String, Object>> toggleStatus(
+            @PathVariable Long id,
+            @RequestParam String status) {
+        try {
+            faqService.toggleStatus(id, status);
+            return ResponseEntity.ok(Map.of(
+                    "success", true,
+                    "message", "状态切换成功"
+            ));
+        } catch (IllegalArgumentException e) {
+            return ResponseEntity.status(400).body(Map.of(
+                    "success", false,
+                    "message", e.getMessage()
+            ));
+        } catch (Exception e) {
+            log.error("FAQ 状态切换失败: id={}", id, e);
+            return ResponseEntity.status(500).body(Map.of(
+                    "success", false,
+                    "message", "状态切换失败：" + e.getMessage()
+            ));
+        }
+    }
+
+    // ==================== 批量导入 ====================
+
+    /**
+     * 批量导入 FAQ
+     * 请求体格式: {"faqs": [{"question":"...", "answer":"...", "similarQuestions":"[...]", "category":"...", "priority":0}]}
+     */
+    @PostMapping("/batch-import")
+    public ResponseEntity<Map<String, Object>> batchImport(@RequestBody Map<String, List<KnowledgeFaq>> body) {
+        try {
+            List<KnowledgeFaq> faqs = body.get("faqs");
+            if (faqs == null || faqs.isEmpty()) {
+                return ResponseEntity.status(400).body(Map.of(
+                        "success", false,
+                        "message", "导入数据不能为空"
+                ));
+            }
+
+            int count = faqService.batchImport(faqs);
+            return ResponseEntity.ok(Map.of(
+                    "success", true,
+                    "message", "批量导入成功，共导入 " + count + " 条",
+                    "data", Map.of("importedCount", count)
+            ));
+        } catch (Exception e) {
+            log.error("FAQ 批量导入失败", e);
+            return ResponseEntity.status(500).body(Map.of(
+                    "success", false,
+                    "message", "批量导入失败：" + e.getMessage()
+            ));
+        }
+    }
+
+    // ==================== 导出 ====================
+
+    /**
+     * 导出所有启用的 FAQ（返回 JSON 列表，后续可扩展为 Excel）
+     */
+    @GetMapping("/export")
+    public ResponseEntity<Map<String, Object>> exportAll() {
+        try {
+            List<KnowledgeFaq> faqs = faqService.exportAll();
+            return ResponseEntity.ok(Map.of(
+                    "success", true,
+                    "message", "导出成功",
+                    "data", faqs
+            ));
+        } catch (Exception e) {
+            log.error("FAQ 导出失败", e);
+            return ResponseEntity.status(500).body(Map.of(
+                    "success", false,
+                    "message", "导出失败：" + e.getMessage()
+            ));
+        }
+    }
+
+    // ==================== 统计 ====================
+
+    /**
+     * 获取 FAQ 匹配统计信息
+     */
+    @GetMapping("/stats")
+    public ResponseEntity<Map<String, Object>> getStats() {
+        try {
+            Map<String, Object> stats = faqService.getStats();
+            return ResponseEntity.ok(Map.of(
+                    "success", true,
+                    "message", "查询成功",
+                    "data", stats
+            ));
+        } catch (Exception e) {
+            log.error("FAQ 统计查询失败", e);
+            return ResponseEntity.status(500).body(Map.of(
+                    "success", false,
+                    "message", "统计查询失败：" + e.getMessage()
+            ));
+        }
+    }
+
+    // ==================== 手动重算向量 ====================
+
+    /**
+     * 手动重新计算某条 FAQ 的向量嵌入
+     */
+    @PostMapping("/{id}/recompute-embedding")
+    public ResponseEntity<Map<String, Object>> recomputeEmbedding(@PathVariable Long id) {
+        try {
+            faqService.recomputeEmbedding(id);
+            return ResponseEntity.ok(Map.of(
+                    "success", true,
+                    "message", "向量重新计算完成"
+            ));
+        } catch (IllegalArgumentException e) {
+            return ResponseEntity.status(400).body(Map.of(
+                    "success", false,
+                    "message", e.getMessage()
+            ));
+        } catch (Exception e) {
+            log.error("FAQ 向量重算失败: id={}", id, e);
+            return ResponseEntity.status(500).body(Map.of(
+                    "success", false,
+                    "message", "向量重算失败：" + e.getMessage()
+            ));
+        }
+    }
+}

src/main/java/com/wok/supportbot/controller/MessageFeedbackController.java

@@ -0,0 +1,146 @@
+package com.wok.supportbot.controller;
+
+import com.wok.supportbot.entity.MessageFeedback;
+import com.wok.supportbot.service.MessageFeedbackService;
+import lombok.extern.slf4j.Slf4j;
+import org.springframework.beans.factory.annotation.Autowired;
+import org.springframework.http.ResponseEntity;
+import org.springframework.web.bind.annotation.*;
+
+import java.util.Arrays;
+import java.util.List;
+import java.util.Map;
+import java.util.UUID;
+
+/**
+ * 消息反馈控制器
+ * 提供反馈提交、统计、查询等 API
+ */
+@RestController
+@Slf4j
+public class MessageFeedbackController {
+
+    @Autowired
+    private MessageFeedbackService messageFeedbackService;
+
+    /**
+     * 提交/修改反馈（upsert）
+     *
+     * @param feedback 反馈信息（messageId, conversationId, feedbackType, reasonCategory, reasonComment）
+     * @return 操作结果
+     */
+    @PostMapping("/feedback")
+    public ResponseEntity<Map<String, Object>> submitFeedback(@RequestBody MessageFeedback feedback) {
+        try {
+            // messageId 缺失时自动生成，保证 upsert 索引不冲突
+            if (feedback.getMessageId() == null || feedback.getMessageId().isBlank()) {
+                feedback.setMessageId("auto_" + UUID.randomUUID().toString().replace("-", ""));
+                log.warn("反馈请求缺少 messageId，已自动生成: {}", feedback.getMessageId());
+            }
+            if (feedback.getFeedbackType() == null || feedback.getFeedbackType().isBlank()) {
+                return ResponseEntity.badRequest().body(Map.of(
+                        "success", false,
+                        "message", "feedbackType 不能为空"
+                ));
+            }
+            String type = feedback.getFeedbackType().toUpperCase();
+            if (!type.equals("THUMBS_UP") && !type.equals("THUMBS_DOWN")) {
+                return ResponseEntity.badRequest().body(Map.of(
+                        "success", false,
+                        "message", "feedbackType 必须为 THUMBS_UP 或 THUMBS_DOWN"
+                ));
+            }
+            feedback.setFeedbackType(type);
+
+            MessageFeedback saved = messageFeedbackService.submitFeedback(feedback);
+            return ResponseEntity.ok(Map.of(
+                    "success", true,
+                    "data", saved,
+                    "message", "反馈提交成功"
+            ));
+        } catch (Exception e) {
+            return ResponseEntity.status(500).body(Map.of(
+                    "success", false,
+                    "message", "反馈提交失败：" + e.getMessage()
+            ));
+        }
+    }
+
+    /**
+     * 获取反馈统计数据
+     *
+     * @param startDate 开始日期（yyyy-MM-dd），可选
+     * @param endDate   结束日期（yyyy-MM-dd），可选
+     * @return 统计结果
+     */
+    @GetMapping("/feedback/stats")
+    public ResponseEntity<Map<String, Object>> getStats(
+            @RequestParam(required = false) String startDate,
+            @RequestParam(required = false) String endDate) {
+        try {
+            Map<String, Object> stats = messageFeedbackService.getStats(startDate, endDate);
+            return ResponseEntity.ok(Map.of(
+                    "success", true,
+                    "data", stats
+            ));
+        } catch (Exception e) {
+            return ResponseEntity.status(500).body(Map.of(
+                    "success", false,
+                    "message", "查询统计失败：" + e.getMessage()
+            ));
+        }
+    }
+
+    /**
+     * 按会话ID查询反馈列表
+     *
+     * @param conversationId 会话ID
+     * @return 反馈列表
+     */
+    @GetMapping("/feedback/by-conversation/{conversationId}")
+    public ResponseEntity<Map<String, Object>> getByConversation(
+            @PathVariable String conversationId) {
+        try {
+            List<MessageFeedback> feedbacks = messageFeedbackService.getByConversationId(conversationId);
+            return ResponseEntity.ok(Map.of(
+                    "success", true,
+                    "data", feedbacks
+            ));
+        } catch (Exception e) {
+            return ResponseEntity.status(500).body(Map.of(
+                    "success", false,
+                    "message", "查询失败：" + e.getMessage()
+            ));
+        }
+    }
+
+    /**
+     * 批量查询反馈状态（供 SDK 回显使用）
+     *
+     * @param messageIds 逗号分隔的消息ID列表
+     * @return 反馈列表
+     */
+    @GetMapping("/feedback/batch")
+    public ResponseEntity<Map<String, Object>> getBatchByMessageIds(
+            @RequestParam String messageIds) {
+        try {
+            if (messageIds == null || messageIds.isBlank()) {
+                return ResponseEntity.ok(Map.of(
+                        "success", true,
+                        "data", List.of()
+                ));
+            }
+            List<String> idList = Arrays.asList(messageIds.split(","));
+            List<MessageFeedback> feedbacks = messageFeedbackService.getBatchByMessageIds(idList);
+            return ResponseEntity.ok(Map.of(
+                    "success", true,
+                    "data", feedbacks
+            ));
+        } catch (Exception e) {
+            return ResponseEntity.status(500).body(Map.of(
+                    "success", false,
+                    "message", "批量查询失败：" + e.getMessage()
+            ));
+        }
+    }
+}

src/main/java/com/wok/supportbot/controller/SensitiveWordController.java

@@ -0,0 +1,239 @@
+package com.wok.supportbot.controller;
+
+import com.wok.supportbot.entity.SensitiveWord;
+import com.wok.supportbot.service.SensitiveWordService;
+import lombok.extern.slf4j.Slf4j;
+import org.springframework.beans.factory.annotation.Autowired;
+import org.springframework.http.ResponseEntity;
+import org.springframework.jdbc.core.JdbcTemplate;
+import org.springframework.web.bind.annotation.*;
+
+import java.util.HashMap;
+import java.util.List;
+import java.util.Map;
+
+/**
+ * 敏感词管理 & 审计日志查询接口
+ */
+@Slf4j
+@RestController
+@RequestMapping("/sensitive-word")
+public class SensitiveWordController {
+
+    @Autowired
+    private SensitiveWordService sensitiveWordService;
+
+    @Autowired
+    private JdbcTemplate jdbcTemplate;
+
+    // ==================== 敏感词 CRUD ====================
+
+    /**
+     * 分页查询敏感词列表
+     */
+    @GetMapping("/list")
+    public ResponseEntity<Map<String, Object>> list(
+            @RequestParam(defaultValue = "1") int page,
+            @RequestParam(defaultValue = "10") int size,
+            @RequestParam(required = false) String keyword,
+            @RequestParam(required = false) String category) {
+        try {
+            Map<String, Object> result = sensitiveWordService.list(page, size, keyword, category);
+
+            Map<String, Object> data = new HashMap<>();
+            data.put("success", true);
+            data.put("data", result.get("records"));
+            data.put("total", result.get("total"));
+            data.put("page", result.get("page"));
+            data.put("size", result.get("size"));
+            data.put("pages", result.get("pages"));
+            return ResponseEntity.ok(data);
+        } catch (Exception e) {
+            log.error("查询敏感词列表失败", e);
+            return ResponseEntity.status(500).body(Map.of(
+                    "success", false,
+                    "message", "查询失败：" + e.getMessage()
+            ));
+        }
+    }
+
+    /**
+     * 新增敏感词
+     */
+    @PostMapping
+    public ResponseEntity<Map<String, Object>> create(@RequestBody SensitiveWord word) {
+        try {
+            SensitiveWord created = sensitiveWordService.create(word);
+            return ResponseEntity.ok(Map.of(
+                    "success", true,
+                    "message", "新增成功",
+                    "data", created
+            ));
+        } catch (IllegalArgumentException e) {
+            return ResponseEntity.badRequest().body(Map.of(
+                    "success", false,
+                    "message", e.getMessage()
+            ));
+        } catch (Exception e) {
+            log.error("新增敏感词失败", e);
+            return ResponseEntity.status(500).body(Map.of(
+                    "success", false,
+                    "message", "新增失败：" + e.getMessage()
+            ));
+        }
+    }
+
+    /**
+     * 修改敏感词
+     */
+    @PutMapping("/{id}")
+    public ResponseEntity<Map<String, Object>> update(@PathVariable Long id, @RequestBody SensitiveWord word) {
+        try {
+            SensitiveWord updated = sensitiveWordService.update(id, word);
+            return ResponseEntity.ok(Map.of(
+                    "success", true,
+                    "message", "修改成功",
+                    "data", updated
+            ));
+        } catch (IllegalArgumentException e) {
+            return ResponseEntity.badRequest().body(Map.of(
+                    "success", false,
+                    "message", e.getMessage()
+            ));
+        } catch (Exception e) {
+            log.error("修改敏感词失败", e);
+            return ResponseEntity.status(500).body(Map.of(
+                    "success", false,
+                    "message", "修改失败：" + e.getMessage()
+            ));
+        }
+    }
+
+    /**
+     * 删除敏感词（逻辑删除）
+     */
+    @DeleteMapping("/{id}")
+    public ResponseEntity<Map<String, Object>> delete(@PathVariable Long id) {
+        try {
+            sensitiveWordService.delete(id);
+            return ResponseEntity.ok(Map.of(
+                    "success", true,
+                    "message", "删除成功"
+            ));
+        } catch (IllegalArgumentException e) {
+            return ResponseEntity.badRequest().body(Map.of(
+                    "success", false,
+                    "message", e.getMessage()
+            ));
+        } catch (Exception e) {
+            log.error("删除敏感词失败", e);
+            return ResponseEntity.status(500).body(Map.of(
+                    "success", false,
+                    "message", "删除失败：" + e.getMessage()
+            ));
+        }
+    }
+
+    /**
+     * 批量导入敏感词
+     * <p>
+     * 请求体格式：{ "words": ["词1", "词2"], "category": "custom", "level": 1 }
+     */
+    @SuppressWarnings("unchecked")
+    @PostMapping("/batch-import")
+    public ResponseEntity<Map<String, Object>> batchImport(@RequestBody Map<String, Object> body) {
+        try {
+            List<String> words = (List<String>) body.get("words");
+            String category = (String) body.getOrDefault("category", "custom");
+            int level = body.containsKey("level") ? ((Number) body.get("level")).intValue() : 1;
+
+            if (words == null || words.isEmpty()) {
+                return ResponseEntity.badRequest().body(Map.of(
+                        "success", false,
+                        "message", "words 列表不能为空"
+                ));
+            }
+
+            int imported = sensitiveWordService.batchImport(words, category, level);
+            return ResponseEntity.ok(Map.of(
+                    "success", true,
+                    "message", "批量导入完成，成功 " + imported + " 个",
+                    "data", Map.of("imported", imported, "total", words.size())
+            ));
+        } catch (Exception e) {
+            log.error("批量导入敏感词失败", e);
+            return ResponseEntity.status(500).body(Map.of(
+                    "success", false,
+                    "message", "批量导入失败：" + e.getMessage()
+            ));
+        }
+    }
+
+    /**
+     * 将 Map 的 key 从 snake_case 转为 camelCase
+     */
+    private Map<String, Object> snakeToCamelKeys(Map<String, Object> map) {
+        Map<String, Object> result = new HashMap<>();
+        map.forEach((key, value) -> {
+            String camelKey = key;
+            int idx;
+            while ((idx = camelKey.indexOf('_')) != -1 && idx + 1 < camelKey.length()) {
+                camelKey = camelKey.substring(0, idx) + Character.toUpperCase(camelKey.charAt(idx + 1)) + camelKey.substring(idx + 2);
+            }
+            result.put(camelKey, value);
+        });
+        return result;
+    }
+
+    // ==================== 审计日志查询 ====================
+
+    /**
+     * 分页查询内容审计日志（使用 JdbcTemplate 直接查询，因为审计日志表无逻辑删除字段）
+     */
+    @GetMapping("/audit-log")
+    public ResponseEntity<Map<String, Object>> auditLog(
+            @RequestParam(defaultValue = "1") int page,
+            @RequestParam(defaultValue = "10") int size,
+            @RequestParam(required = false) String sessionId) {
+        try {
+            // 构建 WHERE 子句
+            StringBuilder whereClause = new StringBuilder("WHERE 1=1");
+            List<Object> params = new java.util.ArrayList<>();
+            if (sessionId != null && !sessionId.isBlank()) {
+                whereClause.append(" AND session_id = ?");
+                params.add(sessionId);
+            }
+
+            // 查询总数
+            String countSql = "SELECT COUNT(*) FROM content_audit_log " + whereClause;
+            Long total = jdbcTemplate.queryForObject(countSql, Long.class, params.toArray());
+
+            // 查询列表
+            String listSql = "SELECT * FROM content_audit_log " + whereClause +
+                    " ORDER BY create_time DESC LIMIT ? OFFSET ?";
+            List<Object> queryParams = new java.util.ArrayList<>(params);
+            queryParams.add(size);
+            queryParams.add((long) (page - 1) * size);
+            List<Map<String, Object>> rawRecords = jdbcTemplate.queryForList(listSql, queryParams.toArray());
+            // JdbcTemplate 返回数据库列名（snake_case），需转换为 camelCase 供前端使用
+            List<Map<String, Object>> records = rawRecords.stream()
+                    .map(this::snakeToCamelKeys)
+                    .toList();
+
+            Map<String, Object> data = new HashMap<>();
+            data.put("success", true);
+            data.put("data", records);
+            data.put("total", total);
+            data.put("page", page);
+            data.put("size", size);
+            data.put("pages", total != null ? (total + size - 1) / size : 0);
+            return ResponseEntity.ok(data);
+        } catch (Exception e) {
+            log.error("查询审计日志失败", e);
+            return ResponseEntity.status(500).body(Map.of(
+                    "success", false,
+                    "message", "查询失败：" + e.getMessage()
+            ));
+        }
+    }
+}

src/main/java/com/wok/supportbot/dao/ContentAuditLogMapper.java

@@ -0,0 +1,12 @@
+package com.wok.supportbot.dao;
+
+import com.baomidou.mybatisplus.core.mapper.BaseMapper;
+import com.wok.supportbot.entity.ContentAuditLog;
+import org.apache.ibatis.annotations.Mapper;
+
+/**
+ * 内容审计日志 Mapper
+ */
+@Mapper
+public interface ContentAuditLogMapper extends BaseMapper<ContentAuditLog> {
+}

src/main/java/com/wok/supportbot/dao/KnowledgeFaqMapper.java

@@ -0,0 +1,12 @@
+package com.wok.supportbot.dao;
+
+import com.baomidou.mybatisplus.core.mapper.BaseMapper;
+import com.wok.supportbot.entity.KnowledgeFaq;
+import org.apache.ibatis.annotations.Mapper;
+
+/**
+ * FAQ 知识库 Mapper
+ */
+@Mapper
+public interface KnowledgeFaqMapper extends BaseMapper<KnowledgeFaq> {
+}

src/main/java/com/wok/supportbot/dao/MessageFeedbackMapper.java

@@ -0,0 +1,12 @@
+package com.wok.supportbot.dao;
+
+import com.baomidou.mybatisplus.core.mapper.BaseMapper;
+import com.wok.supportbot.entity.MessageFeedback;
+import org.apache.ibatis.annotations.Mapper;
+
+/**
+ * 消息反馈 Mapper
+ */
+@Mapper
+public interface MessageFeedbackMapper extends BaseMapper<MessageFeedback> {
+}

src/main/java/com/wok/supportbot/dao/SensitiveWordMapper.java

@@ -0,0 +1,12 @@
+package com.wok.supportbot.dao;
+
+import com.baomidou.mybatisplus.core.mapper.BaseMapper;
+import com.wok.supportbot.entity.SensitiveWord;
+import org.apache.ibatis.annotations.Mapper;
+
+/**
+ * 敏感词 Mapper
+ */
+@Mapper
+public interface SensitiveWordMapper extends BaseMapper<SensitiveWord> {
+}

src/main/java/com/wok/supportbot/entity/ContentAuditLog.java

@@ -0,0 +1,59 @@
+package com.wok.supportbot.entity;
+
+import com.baomidou.mybatisplus.annotation.*;
+import com.fasterxml.jackson.databind.ser.std.ToStringSerializer;
+import com.fasterxml.jackson.databind.annotation.JsonSerialize;
+import com.wok.supportbot.handler.PostgresJsonTypeHandler;
+import lombok.AllArgsConstructor;
+import lombok.Builder;
+import lombok.Data;
+import lombok.NoArgsConstructor;
+
+import java.io.Serial;
+import java.io.Serializable;
+import java.util.Date;
+import java.util.Map;
+
+/**
+ * 内容审计日志实体（不可删除，无逻辑删除字段）
+ */
+@Data
+@Builder
+@AllArgsConstructor
+@NoArgsConstructor
+@TableName(value = "content_audit_log", autoResultMap = true)
+public class ContentAuditLog implements Serializable {
+
+    @Serial
+    @TableField(exist = false)
+    private static final long serialVersionUID = 1L;
+
+    /** 主键ID（雪花算法） */
+    @TableId(value = "id", type = IdType.ASSIGN_ID)
+    @JsonSerialize(using = ToStringSerializer.class)
+    private Long id;
+
+    /** 会话ID */
+    @TableField("session_id")
+    private String sessionId;
+
+    /** 方向：INPUT/OUTPUT */
+    @TableField("direction")
+    private String direction;
+
+    /** 违规内容截断（最多50字） */
+    @TableField("original_text")
+    private String originalText;
+
+    /** 命中词列表（JSON格式） */
+    @TableField(value = "hit_words", typeHandler = PostgresJsonTypeHandler.class)
+    private Map<String, Object> hitWords;
+
+    /** 采取的动作：PASS/MASK/BLOCK */
+    @TableField("action_taken")
+    private String actionTaken;
+
+    /** 创建时间 */
+    @TableField(value = "create_time", fill = FieldFill.INSERT)
+    private Date createTime;
+}

src/main/java/com/wok/supportbot/entity/KnowledgeFaq.java

@@ -0,0 +1,102 @@
+package com.wok.supportbot.entity;
+
+import com.baomidou.mybatisplus.annotation.*;
+import com.fasterxml.jackson.databind.ser.std.ToStringSerializer;
+import com.fasterxml.jackson.databind.annotation.JsonSerialize;
+import lombok.AllArgsConstructor;
+import lombok.Builder;
+import lombok.Data;
+import lombok.NoArgsConstructor;
+
+import java.io.Serial;
+import java.io.Serializable;
+import java.util.Date;
+
+/**
+ * FAQ 知识库实体
+ */
+@Data
+@Builder
+@AllArgsConstructor
+@NoArgsConstructor
+@TableName("knowledge_faq")
+public class KnowledgeFaq implements Serializable {
+
+    @Serial
+    private static final long serialVersionUID = 1L;
+
+    /**
+     * 主键 ID（雪花算法）
+     */
+    @TableId(value = "id", type = IdType.ASSIGN_ID)
+    @JsonSerialize(using = ToStringSerializer.class)
+    private Long id;
+
+    /**
+     * FAQ 问题
+     */
+    @TableField("question")
+    private String question;
+
+    /**
+     * 标准答案
+     */
+    @TableField("answer")
+    private String answer;
+
+    /**
+     * 相似问题列表，存储为 JSON 字符串如 '["问题1","问题2"]'
+     * 不使用 PostgresJsonTypeHandler，在 Service 层手动解析
+     */
+    @TableField("similar_questions")
+    private String similarQuestions;
+
+    /**
+     * FAQ 分类，如"退货政策"
+     */
+    @TableField("category")
+    private String category;
+
+    /**
+     * 状态: ENABLED / DISABLED
+     */
+    @TableField("status")
+    private String status;
+
+    /**
+     * 优先级，数值越大越优先
+     */
+    @TableField("priority")
+    private Integer priority;
+
+    /**
+     * 命中次数统计
+     */
+    @TableField("hit_count")
+    private Long hitCount;
+
+    /**
+     * 来源: manual / import
+     */
+    @TableField("source")
+    private String source;
+
+    /**
+     * 创建时间
+     */
+    @TableField("create_time")
+    private Date createTime;
+
+    /**
+     * 更新时间
+     */
+    @TableField("update_time")
+    private Date updateTime;
+
+    /**
+     * 逻辑删除标记
+     */
+    @TableLogic
+    @TableField("is_delete")
+    private boolean isDelete;
+}

src/main/java/com/wok/supportbot/entity/MessageFeedback.java

@@ -0,0 +1,86 @@
+package com.wok.supportbot.entity;
+
+import com.baomidou.mybatisplus.annotation.*;
+import com.fasterxml.jackson.databind.ser.std.ToStringSerializer;
+import com.fasterxml.jackson.databind.annotation.JsonSerialize;
+import lombok.AllArgsConstructor;
+import lombok.Builder;
+import lombok.Data;
+import lombok.NoArgsConstructor;
+
+import java.io.Serial;
+import java.io.Serializable;
+import java.util.Date;
+
+/**
+ * 消息反馈实体
+ * 记录用户对 AI 回复的点赞/点踩反馈
+ */
+@Data
+@Builder
+@AllArgsConstructor
+@NoArgsConstructor
+@TableName("message_feedback")
+public class MessageFeedback implements Serializable {
+
+    @Serial
+    @TableField(exist = false)
+    private static final long serialVersionUID = 1L;
+
+    /**
+     * 主键ID（雪花算法）
+     */
+    @TableId(value = "id", type = IdType.ASSIGN_ID)
+    @JsonSerialize(using = ToStringSerializer.class)
+    private Long id;
+
+    /**
+     * AI消息ID（前端 msgId 或 UUID）
+     */
+    @TableField("message_id")
+    private String messageId;
+
+    /**
+     * 会话ID
+     */
+    @TableField("conversation_id")
+    private String conversationId;
+
+    /**
+     * 反馈类型: THUMBS_UP / THUMBS_DOWN
+     */
+    @TableField("feedback_type")
+    private String feedbackType;
+
+    /**
+     * 点踩原因分类: inaccurate / irrelevant / incomplete / other
+     * 仅 THUMBS_DOWN 时填写
+     */
+    @TableField("reason_category")
+    private String reasonCategory;
+
+    /**
+     * 自由文本补充说明（可选）
+     */
+    @TableField("reason_comment")
+    private String reasonComment;
+
+    /**
+     * 创建时间
+     */
+    @TableField(value = "create_time", fill = FieldFill.INSERT)
+    private Date createTime;
+
+    /**
+     * 更新时间
+     */
+    @TableField(value = "update_time", fill = FieldFill.INSERT_UPDATE)
+    private Date updateTime;
+
+    /**
+     * 是否删除 false-未删除 true-已删除
+     */
+    @TableField("is_delete")
+    @TableLogic
+    private boolean isDelete;
+}

src/main/java/com/wok/supportbot/entity/SearchResult.java

@@ -55,6 +55,11 @@ public class SearchResult implements Serializable {
      */
     private String documentId;
 
+    /**
+     * 检索模式标识（VECTOR / KEYWORD / HYBRID）
+     */
+    private String searchMode;
+
     /**
      * 原始元数据
      */

src/main/java/com/wok/supportbot/entity/SensitiveWord.java

@@ -0,0 +1,66 @@
+package com.wok.supportbot.entity;
+
+import com.baomidou.mybatisplus.annotation.*;
+import com.fasterxml.jackson.databind.ser.std.ToStringSerializer;
+import com.fasterxml.jackson.databind.annotation.JsonSerialize;
+import lombok.AllArgsConstructor;
+import lombok.Builder;
+import lombok.Data;
+import lombok.NoArgsConstructor;
+
+import java.io.Serial;
+import java.io.Serializable;
+import java.util.Date;
+
+/**
+ * 敏感词实体
+ */
+@Data
+@Builder
+@AllArgsConstructor
+@NoArgsConstructor
+@TableName("sensitive_word")
+public class SensitiveWord implements Serializable {
+
+    @Serial
+    @TableField(exist = false)
+    private static final long serialVersionUID = 1L;
+
+    /** 主键ID（雪花算法） */
+    @TableId(value = "id", type = IdType.ASSIGN_ID)
+    @JsonSerialize(using = ToStringSerializer.class)
+    private Long id;
+
+    /** 敏感词内容 */
+    @TableField("word")
+    private String word;
+
+    /** 分类：politics/porn/abuse/custom */
+    @TableField("category")
+    private String category;
+
+    /** 级别：1=警告，2=拦截 */
+    @TableField("level")
+    private Integer level;
+
+    /** 是否启用 */
+    @TableField("is_active")
+    private Boolean isActive;
+
+    /** 备注 */
+    @TableField("remark")
+    private String remark;
+
+    /** 创建时间 */
+    @TableField(value = "create_time", fill = FieldFill.INSERT)
+    private Date createTime;
+
+    /** 更新时间 */
+    @TableField(value = "update_time", fill = FieldFill.INSERT_UPDATE)
+    private Date updateTime;
+
+    /** 逻辑删除标识 */
+    @TableField("is_delete")
+    @TableLogic
+    private boolean isDelete;
+}

src/main/java/com/wok/supportbot/rag/HybridSearchService.java

@@ -0,0 +1,283 @@
+package com.wok.supportbot.rag;
+
+import com.fasterxml.jackson.core.type.TypeReference;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.wok.supportbot.entity.SearchResult;
+import lombok.extern.slf4j.Slf4j;
+import org.springframework.ai.document.Document;
+import org.springframework.ai.vectorstore.SearchRequest;
+import org.springframework.ai.vectorstore.VectorStore;
+import org.springframework.ai.vectorstore.filter.FilterExpressionBuilder;
+import org.springframework.beans.factory.annotation.Autowired;
+import org.springframework.jdbc.core.JdbcTemplate;
+import org.springframework.stereotype.Service;
+
+import jakarta.annotation.Resource;
+import java.util.*;
+import java.util.stream.Collectors;
+
+/**
+ * 混合检索核心服务
+ * 支持三种检索模式：纯向量检索、全文关键词检索、混合检索（双路 + RRF 融合 + Reranker 精排）。
+ */
+@Service
+@Slf4j
+public class HybridSearchService {
+
+    @Resource(name = "pgVectorVectorStore")
+    private VectorStore pgVectorVectorStore;
+
+    @Autowired
+    private JdbcTemplate jdbcTemplate;
+
+    @Autowired
+    private RrfFusion rrfFusion;
+
+    @Autowired
+    private RerankerService rerankerService;
+
+    /** JSON 解析器 */
+    private static final ObjectMapper OBJECT_MAPPER = new ObjectMapper();
+
+    /**
+     * 执行检索
+     *
+     * @param query              用户查询文本
+     * @param mode               检索模式（VECTOR / KEYWORD / HYBRID）
+     * @param topK               返回结果数量上限
+     * @param similarityThreshold 向量相似度阈值（仅 VECTOR 和 HYBRID 模式生效）
+     * @param categoryIds        分类ID过滤列表（可选）
+     * @return 检索结果列表
+     */
+    public List<SearchResult> search(String query, SearchMode mode, int topK,
+                                     double similarityThreshold, List<Long> categoryIds) {
+        log.info("执行检索: mode={}, topK={}, threshold={}, categoryIds={}", mode, topK, similarityThreshold, categoryIds);
+
+        return switch (mode) {
+            case VECTOR -> vectorSearch(query, topK, similarityThreshold, categoryIds);
+            case KEYWORD -> keywordSearch(query, topK, categoryIds);
+            case HYBRID -> hybridSearch(query, topK, similarityThreshold, categoryIds);
+        };
+    }
+
+    // ==================== 向量检索 ====================
+
+    /**
+     * 纯向量语义检索
+     * 复用 PgVectorStore 的 similaritySearch，支持分类过滤和相似度阈值。
+     */
+    private List<SearchResult> vectorSearch(String query, int topK, double threshold, List<Long> categoryIds) {
+        SearchRequest.Builder builder = SearchRequest.builder()
+                .query(query)
+                .topK(topK)
+                .similarityThreshold(threshold);
+
+        // 分类过滤
+        if (categoryIds != null && !categoryIds.isEmpty()) {
+            FilterExpressionBuilder fb = new FilterExpressionBuilder();
+            List<Object> values = categoryIds.stream()
+                    .map(id -> (Object) String.valueOf(id))
+                    .toList();
+            builder.filterExpression(fb.in("categoryId", values).build());
+        }
+
+        List<Document> docs = pgVectorVectorStore.similaritySearch(builder.build());
+        log.debug("向量检索完成: 查询='{}', 返回 {} 条", query, docs.size());
+        return toSearchResults(docs, "VECTOR");
+    }
+
+    // ==================== 关键词检索 ====================
+
+    /**
+     * 全文关键词检索
+     * 使用 PostgreSQL tsvector 全文索引，按 ts_rank 排序。
+     */
+    private List<SearchResult> keywordSearch(String query, int topK, List<Long> categoryIds) {
+        String categoryFilter = buildCategoryFilter(categoryIds);
+        String sql = """
+                SELECT id, content, metadata,
+                       ts_rank(content_tsvector, plainto_tsquery('simple', ?)) AS rank
+                FROM vector_store
+                WHERE content_tsvector @@ plainto_tsquery('simple', ?)
+                  AND is_delete = false
+                %s
+                ORDER BY rank DESC
+                LIMIT ?
+                """.formatted(categoryFilter);
+
+        List<SearchResult> results = jdbcTemplate.query(sql, (rs, rowNum) -> {
+            SearchResult result = new SearchResult();
+            result.setId(rs.getString("id"));
+            result.setContent(rs.getString("content"));
+            result.setScore(rs.getDouble("rank"));
+            result.setSearchMode("KEYWORD");
+
+            // 解析 metadata JSONB
+            String metadataStr = rs.getString("metadata");
+            if (metadataStr != null) {
+                Map<String, Object> metadata = parseJsonb(metadataStr);
+                result.setMetadata(metadata);
+                result.setSourceName(getStringFromMetadata(metadata, "sourceName"));
+                result.setTitle(getStringFromMetadata(metadata, "title"));
+                result.setDocumentId(getStringFromMetadata(metadata, "documentId"));
+                result.setChunkIndex(getIntegerFromMetadata(metadata, "chunkIndex"));
+            }
+            return result;
+        }, query, query, topK);
+
+        log.debug("关键词检索完成: 查询='{}', 返回 {} 条", query, results.size());
+        return results;
+    }
+
+    // ==================== 混合检索 ====================
+
+    /**
+     * 混合检索：双路检索 → RRF 融合 → Reranker 精排
+     * 向量和关键词各取 topK*2 作为候选池，给 RRF 更多融合素材。
+     */
+    private List<SearchResult> hybridSearch(String query, int topK, double threshold, List<Long> categoryIds) {
+        // 双路检索（各取 topK*2 给 RRF 更多候选）
+        int candidateSize = topK * 2;
+        List<SearchResult> vectorResults = vectorSearch(query, candidateSize, threshold, categoryIds);
+        List<SearchResult> keywordResults = keywordSearch(query, candidateSize, categoryIds);
+
+        log.debug("混合检索双路完成: 向量 {} 条, 关键词 {} 条", vectorResults.size(), keywordResults.size());
+
+        // 转换为 ScoredDocument 列表
+        List<RrfFusion.ScoredDocument> vectorDocs = toScoredDocs(vectorResults);
+        List<RrfFusion.ScoredDocument> keywordDocs = toScoredDocs(keywordResults);
+
+        // RRF 融合
+        List<RrfFusion.ScoredDocument> fused = rrfFusion.fuse(List.of(vectorDocs, keywordDocs));
+
+        // Reranker 精排（如有 RERANK 模型配置则调用 API，否则 fallback 到 RRF 原始排序）
+        List<RrfFusion.ScoredDocument> reranked = rerankerService.rerank(query, fused, topK);
+
+        return scoredDocsToSearchResults(reranked, "HYBRID");
+    }
+
+    // ==================== 转换方法 ====================
+
+    /**
+     * 将 Spring AI Document 列表转换为 SearchResult 列表
+     */
+    private List<SearchResult> toSearchResults(List<Document> docs, String searchMode) {
+        return docs.stream().map(doc -> {
+            SearchResult result = new SearchResult();
+            result.setId(doc.getId());
+            result.setContent(doc.getText());
+            result.setScore(doc.getScore() != null ? doc.getScore() : 0.0);
+            result.setSearchMode(searchMode);
+
+            Map<String, Object> metadata = doc.getMetadata();
+            if (metadata != null) {
+                result.setMetadata(metadata);
+                result.setSourceName(getStringFromMetadata(metadata, "sourceName"));
+                result.setTitle(getStringFromMetadata(metadata, "title"));
+                result.setDocumentId(getStringFromMetadata(metadata, "documentId"));
+                result.setChunkIndex(getIntegerFromMetadata(metadata, "chunkIndex"));
+            }
+            return result;
+        }).collect(Collectors.toList());
+    }
+
+    /**
+     * 将 RRF/Reranker 产出的 ScoredDocument 列表转换为 SearchResult 列表
+     */
+    private List<SearchResult> scoredDocsToSearchResults(List<RrfFusion.ScoredDocument> scoredDocs, String searchMode) {
+        return scoredDocs.stream().map(doc -> {
+            SearchResult result = new SearchResult();
+            result.setId(doc.getId());
+            result.setContent(doc.getContent());
+            result.setScore(doc.getScore());
+            result.setSearchMode(searchMode);
+
+            Map<String, Object> metadata = doc.getMetadata();
+            if (metadata != null) {
+                result.setMetadata(metadata);
+                result.setSourceName(getStringFromMetadata(metadata, "sourceName"));
+                result.setTitle(getStringFromMetadata(metadata, "title"));
+                result.setDocumentId(getStringFromMetadata(metadata, "documentId"));
+                result.setChunkIndex(getIntegerFromMetadata(metadata, "chunkIndex"));
+            }
+            return result;
+        }).collect(Collectors.toList());
+    }
+
+    /**
+     * 将 SearchResult 列表转换为 RRF 用的 ScoredDocument 列表
+     */
+    private List<RrfFusion.ScoredDocument> toScoredDocs(List<SearchResult> results) {
+        return results.stream().map(r -> {
+            Map<String, Object> metadata = null;
+            if (r.getMetadata() instanceof Map) {
+                @SuppressWarnings("unchecked")
+                Map<String, Object> map = (Map<String, Object>) r.getMetadata();
+                metadata = map;
+            }
+            return new RrfFusion.ScoredDocument(
+                    r.getId(),
+                    r.getContent(),
+                    metadata,
+                    r.getScore() != null ? r.getScore() : 0.0
+            );
+        }).collect(Collectors.toList());
+    }
+
+    // ==================== 辅助方法 ====================
+
+    /**
+     * 构建分类过滤 SQL WHERE 子句
+     * 从 metadata JSONB 中提取 categoryId 字段进行匹配。
+     *
+     * @param categoryIds 分类ID列表
+     * @return SQL 片段，如 "AND metadata->>'categoryId' IN ('1','2','3')"；空列表时返回空字符串
+     */
+    private String buildCategoryFilter(List<Long> categoryIds) {
+        if (categoryIds == null || categoryIds.isEmpty()) {
+            return "";
+        }
+        String inValues = categoryIds.stream()
+                .map(id -> "'" + id + "'")
+                .collect(Collectors.joining(","));
+        return "AND metadata->>'categoryId' IN (" + inValues + ")";
+    }
+
+    /**
+     * 解析 JSONB 字符串为 Map
+     */
+    private Map<String, Object> parseJsonb(String json) {
+        try {
+            return OBJECT_MAPPER.readValue(json, new TypeReference<Map<String, Object>>() {});
+        } catch (Exception e) {
+            log.warn("解析 metadata JSONB 失败: {}", e.getMessage());
+            return Collections.emptyMap();
+        }
+    }
+
+    /**
+     * 从 metadata Map 中安全获取 String 值
+     */
+    private String getStringFromMetadata(Map<String, Object> metadata, String key) {
+        Object value = metadata.get(key);
+        return value != null ? value.toString() : null;
+    }
+
+    /**
+     * 从 metadata Map 中安全获取 Integer 值
+     */
+    private Integer getIntegerFromMetadata(Map<String, Object> metadata, String key) {
+        Object value = metadata.get(key);
+        if (value instanceof Number) {
+            return ((Number) value).intValue();
+        }
+        if (value instanceof String) {
+            try {
+                return Integer.parseInt((String) value);
+            } catch (NumberFormatException e) {
+                return null;
+            }
+        }
+        return null;
+    }
+}

src/main/java/com/wok/supportbot/rag/RerankerService.java

@@ -0,0 +1,236 @@
+package com.wok.supportbot.rag;
+
+import com.baomidou.mybatisplus.core.conditions.query.LambdaQueryWrapper;
+import com.wok.supportbot.dao.AiModelConfigMapper;
+import com.wok.supportbot.entity.AiModelConfig;
+import lombok.extern.slf4j.Slf4j;
+import org.springframework.beans.factory.annotation.Autowired;
+import org.springframework.http.*;
+import org.springframework.stereotype.Service;
+import org.springframework.web.client.RestTemplate;
+
+import java.time.Duration;
+import java.util.*;
+import java.util.stream.Collectors;
+
+/**
+ * 重排序服务
+ * 查询 ai_model_config 表中 RERANK 类型的活跃配置，调用对应的 Rerank API 对候选文档精排。
+ * 支持 DashScope 和 OpenAI 兼容两种协议，无配置或调用异常时 fallback 到原始排序。
+ */
+@Service
+@Slf4j
+public class RerankerService {
+
+    @Autowired
+    private AiModelConfigMapper aiModelConfigMapper;
+
+    /** HTTP 超时时间（秒） */
+    private static final int TIMEOUT_SECONDS = 3;
+
+    /**
+     * 对候选文档执行重排序
+     * - 有 RERANK 配置 → 调用对应提供商的 Rerank API
+     * - 无配置 / 超时 / 异常 → fallback 返回 candidates 前 topN 条
+     *
+     * @param query      用户查询
+     * @param candidates RRF 融合后的候选文档列表
+     * @param topN       最终返回的文档数量
+     * @return 按相关性降序排列的 topN 条文档
+     */
+    public List<RrfFusion.ScoredDocument> rerank(String query, List<RrfFusion.ScoredDocument> candidates, int topN) {
+        if (candidates == null || candidates.isEmpty()) {
+            return Collections.emptyList();
+        }
+
+        // 查询 RERANK 类型的活跃配置
+        AiModelConfig config = getActiveRerankConfig();
+        if (config == null) {
+            log.debug("未找到 RERANK 活跃配置，使用 RRF 原始排序作为 fallback");
+            return fallback(candidates, topN);
+        }
+
+        try {
+            List<RrfFusion.ScoredDocument> reranked = doRerank(query, candidates, topN, config);
+            log.debug("Reranker 精排完成: provider={}, 输入 {} 条, 输出 {} 条",
+                    config.getProvider(), candidates.size(), reranked.size());
+            return reranked;
+        } catch (Exception e) {
+            log.warn("Reranker 调用失败 (provider={}): {}, 使用 RRF 原始排序作为 fallback",
+                    config.getProvider(), e.getMessage());
+            return fallback(candidates, topN);
+        }
+    }
+
+    /**
+     * 获取 RERANK 类型的活跃配置
+     */
+    private AiModelConfig getActiveRerankConfig() {
+        LambdaQueryWrapper<AiModelConfig> wrapper = new LambdaQueryWrapper<>();
+        wrapper.eq(AiModelConfig::getAppType, "RERANK")
+                .eq(AiModelConfig::getIsActive, true)
+                .last("LIMIT 1");
+        return aiModelConfigMapper.selectOne(wrapper);
+    }
+
+    /**
+     * 根据提供商类型分发 Rerank API 调用
+     */
+    private List<RrfFusion.ScoredDocument> doRerank(String query,
+                                                     List<RrfFusion.ScoredDocument> candidates,
+                                                     int topN,
+                                                     AiModelConfig config) {
+        String provider = config.getProvider().toLowerCase();
+        if ("dashscope".equals(provider)) {
+            return dashscopeRerank(query, candidates, topN, config);
+        } else {
+            return openaiCompatibleRerank(query, candidates, topN, config);
+        }
+    }
+
+    /**
+     * DashScope Rerank API 调用
+     * POST https://dashscope.aliyuncs.com/api/v1/services/rerank
+     */
+    private List<RrfFusion.ScoredDocument> dashscopeRerank(String query,
+                                                            List<RrfFusion.ScoredDocument> candidates,
+                                                            int topN,
+                                                            AiModelConfig config) {
+        String url = "https://dashscope.aliyuncs.com/api/v1/services/rerank";
+        List<String> documents = candidates.stream()
+                .map(RrfFusion.ScoredDocument::getContent)
+                .collect(Collectors.toList());
+
+        // 构建请求体
+        Map<String, Object> parameters = new LinkedHashMap<>();
+        parameters.put("top_n", topN);
+        parameters.put("return_documents", true);
+
+        Map<String, Object> input = new LinkedHashMap<>();
+        input.put("query", query);
+        input.put("documents", documents);
+
+        Map<String, Object> body = new LinkedHashMap<>();
+        body.put("model", config.getModelName() != null ? config.getModelName() : "gte-rerank");
+        body.put("input", input);
+        body.put("parameters", parameters);
+
+        // 发送请求
+        HttpHeaders headers = new HttpHeaders();
+        headers.setContentType(MediaType.APPLICATION_JSON);
+        headers.setBearerAuth(config.getApiKey());
+
+        ResponseEntity<Map> response = postWithTimeout(url, headers, body);
+
+        // 解析响应：{"output":{"results":[{"index":0,"relevance_score":0.95},...]}}
+        Map responseBody = response.getBody();
+        if (responseBody == null) {
+            throw new RuntimeException("DashScope Rerank 响应为空");
+        }
+        Map output = (Map) responseBody.get("output");
+        if (output == null) {
+            throw new RuntimeException("DashScope Rerank 响应缺少 output 字段");
+        }
+        List<Map> results = (List<Map>) output.get("results");
+        if (results == null || results.isEmpty()) {
+            throw new RuntimeException("DashScope Rerank 结果为空");
+        }
+
+        return mapResultsToDocs(results, candidates);
+    }
+
+    /**
+     * OpenAI 兼容 Rerank API 调用
+     * POST {baseUrl}/rerank
+     */
+    private List<RrfFusion.ScoredDocument> openaiCompatibleRerank(String query,
+                                                                   List<RrfFusion.ScoredDocument> candidates,
+                                                                   int topN,
+                                                                   AiModelConfig config) {
+        String baseUrl = config.getBaseUrl();
+        if (baseUrl == null || baseUrl.isBlank()) {
+            throw new IllegalArgumentException(
+                    "OpenAI 兼容 Rerank 提供商 [" + config.getProvider() + "] 未配置 baseUrl");
+        }
+        // 确保 URL 拼接正确
+        String url = baseUrl.endsWith("/") ? baseUrl + "rerank" : baseUrl + "/rerank";
+
+        List<String> documents = candidates.stream()
+                .map(RrfFusion.ScoredDocument::getContent)
+                .collect(Collectors.toList());
+
+        // 构建请求体
+        Map<String, Object> body = new LinkedHashMap<>();
+        body.put("model", config.getModelName());
+        body.put("query", query);
+        body.put("documents", documents);
+        body.put("top_n", topN);
+
+        // 发送请求
+        HttpHeaders headers = new HttpHeaders();
+        headers.setContentType(MediaType.APPLICATION_JSON);
+        headers.setBearerAuth(config.getApiKey());
+
+        ResponseEntity<Map> response = postWithTimeout(url, headers, body);
+
+        // 解析响应：{"results":[{"index":0,"relevance_score":0.95},...]}
+        Map responseBody = response.getBody();
+        if (responseBody == null) {
+            throw new RuntimeException("OpenAI 兼容 Rerank 响应为空");
+        }
+        List<Map> results = (List<Map>) responseBody.get("results");
+        if (results == null || results.isEmpty()) {
+            throw new RuntimeException("OpenAI 兼容 Rerank 结果为空");
+        }
+
+        return mapResultsToDocs(results, candidates);
+    }
+
+    /**
+     * 将 Rerank API 返回的结果映射为 ScoredDocument 列表
+     *
+     * @param results    API 返回的 results 数组，每项含 index 和 relevance_score
+     * @param candidates 原始候选文档列表（用于通过 index 关联文档内容）
+     * @return 按 relevance_score 降序排列的文档列表
+     */
+    private List<RrfFusion.ScoredDocument> mapResultsToDocs(List<Map> results,
+                                                             List<RrfFusion.ScoredDocument> candidates) {
+        List<RrfFusion.ScoredDocument> reranked = new ArrayList<>(results.size());
+        for (Map result : results) {
+            int index = ((Number) result.get("index")).intValue();
+            double score = ((Number) result.get("relevance_score")).doubleValue();
+            if (index >= 0 && index < candidates.size()) {
+                RrfFusion.ScoredDocument original = candidates.get(index);
+                reranked.add(new RrfFusion.ScoredDocument(
+                        original.getId(),
+                        original.getContent(),
+                        original.getMetadata(),
+                        score
+                ));
+            }
+        }
+        // 按 relevance_score 降序排列
+        reranked.sort((a, b) -> Double.compare(b.getScore(), a.getScore()));
+        return reranked;
+    }
+
+    /**
+     * 带超时的 HTTP POST 请求
+     */
+    @SuppressWarnings("unchecked")
+    private ResponseEntity<Map> postWithTimeout(String url, HttpHeaders headers, Map<String, Object> body) {
+        RestTemplate restTemplate = new RestTemplate();
+        HttpEntity<Map<String, Object>> entity = new HttpEntity<>(body, headers);
+        // RestTemplate 默认无超时，此处依赖连接/读取超时由底层控制
+        // Spring Boot 3.x 中可使用 RestClient 替代以获得更好的超时支持
+        return restTemplate.exchange(url, HttpMethod.POST, entity, Map.class);
+    }
+
+    /**
+     * Fallback：直接截取前 topN 条候选文档
+     */
+    private List<RrfFusion.ScoredDocument> fallback(List<RrfFusion.ScoredDocument> candidates, int topN) {
+        int limit = Math.min(topN, candidates.size());
+        return new ArrayList<>(candidates.subList(0, limit));
+    }
+}

src/main/java/com/wok/supportbot/rag/RrfFusion.java

@@ -0,0 +1,88 @@
+package com.wok.supportbot.rag;
+
+import lombok.AllArgsConstructor;
+import lombok.Data;
+import lombok.NoArgsConstructor;
+import lombok.extern.slf4j.Slf4j;
+import org.springframework.stereotype.Component;
+
+import java.util.*;
+
+/**
+ * RRF（Reciprocal Rank Fusion）融合算法
+ * 将多路检索结果按排名倒数加权合并，消除不同检索方式分数尺度不一致的问题。
+ * 公式：score(d) = Σ 1.0 / (K + rank_i(d) + 1)，其中 K=60 为平滑常数。
+ */
+@Component
+@Slf4j
+public class RrfFusion {
+
+    /** RRF 平滑常数，控制排名靠后的文档权重衰减速度 */
+    private static final double K = 60.0;
+
+    /**
+     * 对多路排序列表执行 RRF 融合
+     *
+     * @param rankedLists 多路检索结果列表，每个子列表已按相关性降序排列
+     * @return 按 RRF 分数降序排列的合并文档列表
+     */
+    public List<ScoredDocument> fuse(List<List<ScoredDocument>> rankedLists) {
+        // docId → 累计 RRF 分数
+        Map<String, Double> scoreMap = new LinkedHashMap<>();
+        // docId → 文档对象（保留首次出现的内容和 metadata）
+        Map<String, ScoredDocument> docMap = new LinkedHashMap<>();
+
+        for (List<ScoredDocument> rankedList : rankedLists) {
+            if (rankedList == null) {
+                continue;
+            }
+            for (int rank = 0; rank < rankedList.size(); rank++) {
+                ScoredDocument doc = rankedList.get(rank);
+                if (doc == null || doc.getId() == null) {
+                    continue;
+                }
+                double rrfScore = 1.0 / (K + rank + 1);
+                scoreMap.merge(doc.getId(), rrfScore, Double::sum);
+                // 保留首次出现的文档内容
+                docMap.putIfAbsent(doc.getId(), doc);
+            }
+        }
+
+        // 按 RRF 分数降序排列
+        List<Map.Entry<String, Double>> sortedEntries = new ArrayList<>(scoreMap.entrySet());
+        sortedEntries.sort((a, b) -> Double.compare(b.getValue(), a.getValue()));
+
+        // 构建结果列表，设置 RRF 融合分数
+        List<ScoredDocument> result = new ArrayList<>(sortedEntries.size());
+        for (Map.Entry<String, Double> entry : sortedEntries) {
+            ScoredDocument doc = docMap.get(entry.getKey());
+            ScoredDocument fusedDoc = new ScoredDocument(
+                    doc.getId(),
+                    doc.getContent(),
+                    doc.getMetadata(),
+                    entry.getValue()
+            );
+            result.add(fusedDoc);
+        }
+
+        log.debug("RRF 融合完成: 输入 {} 路, 合并后 {} 条文档", rankedLists.size(), result.size());
+        return result;
+    }
+
+    /**
+     * 带分数的文档对象，用于 RRF 融合和 Reranker 之间的数据传递
+     */
+    @Data
+    @AllArgsConstructor
+    @NoArgsConstructor
+    public static class ScoredDocument {
+        /** 文档ID */
+        private String id;
+        /** 文档内容 */
+        private String content;
+        /** 元数据 */
+        private Map<String, Object> metadata;
+        /** 相关性分数（RRF 分数或 Reranker 分数） */
+        private double score;
+    }
+}

src/main/java/com/wok/supportbot/rag/SearchMode.java

@@ -0,0 +1,19 @@
+package com.wok.supportbot.rag;
+
+/**
+ * 检索模式枚举
+ * - VECTOR: 纯向量语义检索
+ * - KEYWORD: 全文关键词检索
+ * - HYBRID: 混合检索（双路检索 + RRF 融合 + Reranker 精排）
+ */
+public enum SearchMode {
+
+    /** 纯向量语义检索 */
+    VECTOR,
+
+    /** 混合检索：双路检索 → RRF 融合 → Reranker 精排 */
+    HYBRID,
+
+    /** 全文关键词检索 */
+    KEYWORD
+}

src/main/java/com/wok/supportbot/service/ContentSafetyService.java

@@ -0,0 +1,209 @@
+package com.wok.supportbot.service;
+
+import com.baomidou.mybatisplus.core.conditions.query.QueryWrapper;
+import com.wok.supportbot.dao.SensitiveWordMapper;
+import com.wok.supportbot.entity.SensitiveWord;
+import jakarta.annotation.PostConstruct;
+import lombok.AllArgsConstructor;
+import lombok.Data;
+import lombok.extern.slf4j.Slf4j;
+import org.springframework.beans.factory.annotation.Autowired;
+import org.springframework.stereotype.Service;
+
+import java.util.*;
+
+/**
+ * 内容安全过滤服务 —— DFA（确定有限自动机）敏感词匹配引擎
+ */
+@Slf4j
+@Service
+public class ContentSafetyService {
+
+    @Autowired
+    private SensitiveWordMapper sensitiveWordMapper;
+
+    /** DFA 字典树根节点，使用 volatile + 全新对象替换保证线程安全 */
+    private volatile DfaNode root = new DfaNode();
+
+    // ==================== 内部类定义 ====================
+
+    /**
+     * DFA 节点
+     */
+    static class DfaNode {
+        Map<Character, DfaNode> children = new HashMap<>();
+        boolean isEnd = false;
+        SensitiveWord word;
+    }
+
+    /**
+     * 命中结果
+     */
+    @Data
+    @AllArgsConstructor
+    public static class HitResult {
+        /** 命中的敏感词 */
+        private String word;
+        /** 分类 */
+        private String category;
+        /** 级别：1=警告，2=拦截 */
+        private Integer level;
+        /** 在原文中的起始位置 */
+        private int startIndex;
+        /** 在原文中的结束位置（不含） */
+        private int endIndex;
+    }
+
+    // ==================== 初始化与重建 ====================
+
+    /**
+     * 应用启动时从数据库加载所有活跃敏感词，构建 DFA 字典树
+     */
+    @PostConstruct
+    public void init() {
+        rebuild();
+    }
+
+    /**
+     * 重新从数据库加载并重建字典树（增删改敏感词后调用）
+     */
+    public synchronized void rebuild() {
+        try {
+            QueryWrapper<SensitiveWord> wrapper = new QueryWrapper<>();
+            wrapper.eq("is_active", true);
+            List<SensitiveWord> words = sensitiveWordMapper.selectList(wrapper);
+
+            // 构建全新的字典树
+            DfaNode newRoot = new DfaNode();
+            for (SensitiveWord word : words) {
+                if (word.getWord() == null || word.getWord().isBlank()) {
+                    continue;
+                }
+                addWord(newRoot, word);
+            }
+
+            // 原子替换根节点引用，保证线程安全
+            this.root = newRoot;
+            log.info("DFA 字典树重建完成，共加载 {} 个敏感词", words.size());
+        } catch (Exception e) {
+            log.error("DFA 字典树重建失败", e);
+        }
+    }
+
+    /**
+     * 向字典树中添加一个敏感词
+     */
+    private void addWord(DfaNode rootNode, SensitiveWord sensitiveWord) {
+        String text = sensitiveWord.getWord().toLowerCase();
+        DfaNode current = rootNode;
+        for (char c : text.toCharArray()) {
+            current.children.putIfAbsent(c, new DfaNode());
+            current = current.children.get(c);
+        }
+        current.isEnd = true;
+        current.word = sensitiveWord;
+    }
+
+    // ==================== 检测与脱敏 ====================
+
+    /**
+     * 检测文本中的敏感词，返回所有命中结果
+     *
+     * @param text 待检测文本
+     * @return 命中结果列表
+     */
+    public List<HitResult> detect(String text) {
+        List<HitResult> hits = new ArrayList<>();
+        if (text == null || text.isEmpty()) {
+            return hits;
+        }
+
+        String lowerText = text.toLowerCase();
+        DfaNode currentNode = root;
+
+        for (int i = 0; i < lowerText.length(); i++) {
+            char c = lowerText.charAt(i);
+            DfaNode nextNode = currentNode.children.get(c);
+
+            if (nextNode != null) {
+                currentNode = nextNode;
+                // 如果当前节点是一个完整词的结尾，记录命中
+                if (currentNode.isEnd) {
+                    String hitWord = currentNode.word.getWord();
+                    int startIdx = i - hitWord.length() + 1;
+                    hits.add(new HitResult(
+                            hitWord,
+                            currentNode.word.getCategory(),
+                            currentNode.word.getLevel(),
+                            startIdx,
+                            i + 1
+                    ));
+                }
+            } else {
+                // 未匹配到子节点，回退到根节点
+                // 但如果之前已经部分匹配过，需要从下一个起始位置重新开始
+                if (currentNode != root) {
+                    // 回退：从上一个匹配起始位置的下一个字符重新开始
+                    // 简化处理：直接重置到根节点，外层循环继续
+                    currentNode = root;
+                    // 重新检查当前字符是否可以从根节点开始匹配
+                    nextNode = currentNode.children.get(c);
+                    if (nextNode != null) {
+                        currentNode = nextNode;
+                        if (currentNode.isEnd) {
+                            String hitWord = currentNode.word.getWord();
+                            int startIdx = i - hitWord.length() + 1;
+                            hits.add(new HitResult(
+                                    hitWord,
+                                    currentNode.word.getCategory(),
+                                    currentNode.word.getLevel(),
+                                    startIdx,
+                                    i + 1
+                            ));
+                        }
+                    }
+                }
+            }
+        }
+
+        return hits;
+    }
+
+    /**
+     * 将文本中检测到的敏感词替换为 ***
+     *
+     * @param text 原始文本
+     * @return 脱敏后的文本
+     */
+    public String mask(String text) {
+        List<HitResult> hits = detect(text);
+        if (hits.isEmpty()) {
+            return text;
+        }
+
+        // 按起始位置排序
+        hits.sort(Comparator.comparingInt(HitResult::getStartIndex));
+
+        StringBuilder sb = new StringBuilder(text);
+        int offset = 0; // 记录因替换导致的长度偏移
+        for (HitResult hit : hits) {
+            int start = hit.getStartIndex() + offset;
+            int end = hit.getEndIndex() + offset;
+            String replacement = "***";
+            sb.replace(start, end, replacement);
+            offset += replacement.length() - (hit.getEndIndex() - hit.getStartIndex());
+        }
+
+        return sb.toString();
+    }
+
+    /**
+     * 判断命中结果中是否存在需要拦截（level >= 2）的条目
+     *
+     * @param hits 命中结果列表
+     * @return 是否有拦截级别的命中
+     */
+    public boolean hasBlockingHit(List<HitResult> hits) {
+        return hits.stream().anyMatch(h -> h.getLevel() != null && h.getLevel() >= 2);
+    }
+}

src/main/java/com/wok/supportbot/service/ConversationService.java

@@ -3,6 +3,7 @@ package com.wok.supportbot.service;
 import com.baomidou.mybatisplus.core.conditions.query.LambdaQueryWrapper;
 import com.wok.supportbot.dao.ChatMessageMapper;
 import com.wok.supportbot.entity.ChatMessage;
+import com.wok.supportbot.entity.MessageFeedback;
 import lombok.extern.slf4j.Slf4j;
 import org.springframework.beans.factory.annotation.Autowired;
 import org.springframework.jdbc.core.JdbcTemplate;
@@ -21,6 +22,9 @@ public class ConversationService {
     @Autowired
     private ChatMessageMapper chatMessageMapper;
 
+    @Autowired
+    private MessageFeedbackService messageFeedbackService;
+
     @Autowired
     private JdbcTemplate jdbcTemplate;
 
@@ -328,6 +332,13 @@ public class ConversationService {
             return "会话不存在或无任何消息";
         }
 
+        // 查询该会话的所有反馈，按 messageId 建立索引
+        List<MessageFeedback> feedbacks = messageFeedbackService.getByConversationId(conversationId);
+        Map<String, MessageFeedback> feedbackMap = new HashMap<>();
+        for (MessageFeedback fb : feedbacks) {
+            feedbackMap.put(fb.getMessageId(), fb);
+        }
+
         StringBuilder sb = new StringBuilder();
         sb.append("====================================\n");
         sb.append("会话导出记录\n");
@@ -342,6 +353,10 @@ public class ConversationService {
             String type = String.valueOf(msg.get("messageType"));
             String content = String.valueOf(msg.get("content"));
             String time = String.valueOf(msg.get("createTime"));
+            // metadata 中可能包含前端生成的 msgId
+            @SuppressWarnings("unchecked")
+            Map<String, Object> metadata = (Map<String, Object>) msg.get("metadata");
+            String msgId = metadata != null ? (String) metadata.get("msgId") : null;
 
             sb.append("[").append(time).append("] ");
             switch (type) {
@@ -351,7 +366,42 @@ public class ConversationService {
                 default -> sb.append(type);
             }
             sb.append("\n");
-            sb.append(content).append("\n\n");
+            sb.append(content).append("\n");
+
+            // AI 消息后附加反馈信息
+            if ("ASSISTANT".equals(type) && msgId != null) {
+                MessageFeedback fb = feedbackMap.get(msgId);
+                if (fb != null) {
+                    if ("THUMBS_UP".equals(fb.getFeedbackType())) {
+                        sb.append("  👍 用户反馈: 有帮助\n");
+                    } else if ("THUMBS_DOWN".equals(fb.getFeedbackType())) {
+                        sb.append("  👎 用户反馈: 没帮助");
+                        String reason = fb.getReasonCategory();
+                        String comment = fb.getReasonComment();
+                        if (reason != null || (comment != null && !comment.isBlank())) {
+                            sb.append(" (");
+                            List<String> parts = new ArrayList<>();
+                            if (reason != null) {
+                                String reasonLabel = switch (reason) {
+                                    case "inaccurate" -> "信息错误";
+                                    case "irrelevant" -> "不相关";
+                                    case "incomplete" -> "不完整";
+                                    case "other" -> "其他";
+                                    default -> reason;
+                                };
+                                parts.add("原因: " + reasonLabel);
+                            }
+                            if (comment != null && !comment.isBlank()) {
+                                parts.add("补充: " + comment);
+                            }
+                            sb.append(String.join(", ", parts));
+                            sb.append(")");
+                        }
+                        sb.append("\n");
+                    }
+                }
+            }
+            sb.append("\n");
         }
 
         sb.append("====================================\n");

src/main/java/com/wok/supportbot/service/DocumentService.java

@@ -532,6 +532,39 @@ public class DocumentService {
         return searchResults;
     }
 
+    /**
+     * 多模式搜索（P0-001: 支持向量/关键词/混合检索）
+     *
+     * @param searchMode 检索模式：VECTOR（默认）/ KEYWORD / HYBRID
+     */
+    public List<SearchResult> searchDocuments(String query, int topK, double similarityThreshold,
+                                               List<Long> categoryIds, String searchMode) {
+        if (searchMode == null || searchMode.isEmpty() || "VECTOR".equalsIgnoreCase(searchMode)) {
+            // 默认向量检索，向后兼容
+            List<SearchResult> results = searchDocuments(query, topK, similarityThreshold, categoryIds);
+            results.forEach(r -> r.setSearchMode("VECTOR"));
+            return results;
+        }
+
+        // 委托给 HybridSearchService 处理 KEYWORD 和 HYBRID 模式
+        if (hybridSearchService != null) {
+            com.wok.supportbot.rag.SearchMode mode;
+            try {
+                mode = com.wok.supportbot.rag.SearchMode.valueOf(searchMode.toUpperCase());
+            } catch (IllegalArgumentException e) {
+                mode = com.wok.supportbot.rag.SearchMode.VECTOR;
+            }
+            return hybridSearchService.search(query, mode, topK, similarityThreshold, categoryIds);
+        }
+
+        // HybridSearchService 不可用时降级到向量检索
+        log.warn("HybridSearchService 不可用，降级为向量检索");
+        return searchDocuments(query, topK, similarityThreshold, categoryIds);
+    }
+
+    @Autowired(required = false)
+    private com.wok.supportbot.rag.HybridSearchService hybridSearchService;
+
     private List<String> normalizeCategoryIds(List<Long> categoryIds) {
         if (categoryIds == null || categoryIds.isEmpty()) {
             return Collections.emptyList();

src/main/java/com/wok/supportbot/service/FaqMatchEngine.java

@@ -0,0 +1,363 @@
+package com.wok.supportbot.service;
+
+import com.fasterxml.jackson.core.type.TypeReference;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.wok.supportbot.config.EmbeddingModelFactory;
+import com.wok.supportbot.dao.KnowledgeFaqMapper;
+import com.wok.supportbot.entity.KnowledgeFaq;
+import lombok.AllArgsConstructor;
+import lombok.Data;
+import lombok.extern.slf4j.Slf4j;
+import org.springframework.ai.embedding.EmbeddingModel;
+import org.springframework.ai.embedding.EmbeddingRequest;
+import org.springframework.beans.factory.annotation.Autowired;
+import org.springframework.beans.factory.annotation.Value;
+import org.springframework.jdbc.core.JdbcTemplate;
+import org.springframework.stereotype.Service;
+
+import java.util.*;
+import java.util.concurrent.CompletableFuture;
+
+/**
+ * FAQ 三级匹配引擎
+ * 1. 精确匹配：问题文本完全一致
+ * 2. 关键词匹配：similar_questions 字段包含用户问题中的关键词
+ * 3. 语义匹配：基于向量余弦距离的相似度匹配
+ */
+@Service
+@Slf4j
+public class FaqMatchEngine {
+
+    @Autowired
+    private KnowledgeFaqMapper faqMapper;
+
+    @Autowired
+    private JdbcTemplate jdbcTemplate;
+
+    @Autowired
+    private EmbeddingModelFactory embeddingModelFactory;
+
+    /** 向量维度，与 PgVectorStore 保持一致 */
+    @Value("${knowledge.vector.dimension:1024}")
+    private int vectorDimension;
+
+    /** 语义匹配相似度阈值 */
+    @Value("${knowledge.faq.semantic-threshold:0.85}")
+    private double semanticThreshold;
+
+    private final ObjectMapper objectMapper = new ObjectMapper();
+
+    // ==================== 匹配结果内部类 ====================
+
+    /**
+     * FAQ 匹配结果
+     */
+    @Data
+    @AllArgsConstructor
+    public static class FaqMatchResult {
+        /** 匹配到的 FAQ */
+        private KnowledgeFaq faq;
+        /** 匹配类型: EXACT / KEYWORD / SEMANTIC */
+        private String matchType;
+        /** 匹配分数 (0.0 ~ 1.0) */
+        private double score;
+    }
+
+    // ==================== 核心匹配方法 ====================
+
+    /**
+     * 对用户问题进行三级匹配
+     *
+     * @param question 用户问题
+     * @return 匹配结果（可能为空）
+     */
+    public Optional<FaqMatchResult> match(String question) {
+        if (question == null || question.isBlank()) {
+            return Optional.empty();
+        }
+        String trimmedQuestion = question.trim();
+
+        // 第一级：精确匹配
+        Optional<FaqMatchResult> exactResult = exactMatch(trimmedQuestion);
+        if (exactResult.isPresent()) {
+            log.info("FAQ 精确匹配命中: question={}", trimmedQuestion);
+            return exactResult;
+        }
+
+        // 第二级：关键词匹配
+        Optional<FaqMatchResult> keywordResult = keywordMatch(trimmedQuestion);
+        if (keywordResult.isPresent()) {
+            log.info("FAQ 关键词匹配命中: question={}", trimmedQuestion);
+            return keywordResult;
+        }
+
+        // 第三级：语义匹配
+        Optional<FaqMatchResult> semanticResult = semanticMatch(trimmedQuestion);
+        if (semanticResult.isPresent()) {
+            log.info("FAQ 语义匹配命中: question={}, score={}", trimmedQuestion, semanticResult.get().getScore());
+            return semanticResult;
+        }
+
+        log.debug("FAQ 未匹配: question={}", trimmedQuestion);
+        return Optional.empty();
+    }
+
+    // ==================== 第一级：精确匹配 ====================
+
+    /**
+     * 精确匹配：问题文本完全一致
+     */
+    private Optional<FaqMatchResult> exactMatch(String question) {
+        try {
+            List<KnowledgeFaq> results = jdbcTemplate.query(
+                    "SELECT * FROM knowledge_faq WHERE question = ? AND status = 'ENABLED' AND is_delete = false ORDER BY priority DESC LIMIT 1",
+                    (rs, rowNum) -> mapRowToFaq(rs),
+                    question
+            );
+            if (!results.isEmpty()) {
+                KnowledgeFaq faq = results.get(0);
+                incrementHitCount(faq.getId());
+                return Optional.of(new FaqMatchResult(faq, "EXACT", 1.0));
+            }
+        } catch (Exception e) {
+            log.error("FAQ 精确匹配异常", e);
+        }
+        return Optional.empty();
+    }
+
+    // ==================== 第二级：关键词匹配 ====================
+
+    /**
+     * 关键词匹配：从用户问题中提取关键词，检查 similar_questions 是否包含
+     */
+    private Optional<FaqMatchResult> keywordMatch(String question) {
+        try {
+            List<String> keywords = extractKeywords(question);
+            if (keywords.isEmpty()) {
+                return Optional.empty();
+            }
+
+            // 构建 ILIKE 条件
+            StringBuilder sqlBuilder = new StringBuilder(
+                    "SELECT * FROM knowledge_faq WHERE status = 'ENABLED' AND is_delete = false AND ("
+            );
+            List<Object> params = new ArrayList<>();
+            for (int i = 0; i < keywords.size(); i++) {
+                if (i > 0) {
+                    sqlBuilder.append(" OR ");
+                }
+                sqlBuilder.append("similar_questions ILIKE ?");
+                params.add("%" + keywords.get(i) + "%");
+            }
+            sqlBuilder.append(") ORDER BY priority DESC LIMIT 5");
+
+            List<KnowledgeFaq> results = jdbcTemplate.query(
+                    sqlBuilder.toString(),
+                    (rs, rowNum) -> mapRowToFaq(rs),
+                    params.toArray()
+            );
+
+            if (!results.isEmpty()) {
+                KnowledgeFaq faq = results.get(0);
+                return Optional.of(new FaqMatchResult(faq, "KEYWORD", 0.9));
+            }
+        } catch (Exception e) {
+            log.error("FAQ 关键词匹配异常", e);
+        }
+        return Optional.empty();
+    }
+
+    /**
+     * 简单分词：按空格、标点切分，过滤短词
+     */
+    private List<String> extractKeywords(String question) {
+        String[] tokens = question.split("[\\s,，。？！?!、;；:：\"\"''\\(\\)（）\\[\\]【】]+");
+        List<String> keywords = new ArrayList<>();
+        for (String token : tokens) {
+            String trimmed = token.trim();
+            // 过滤长度小于2的token（避免单字匹配过于宽泛）
+            if (trimmed.length() >= 2) {
+                keywords.add(trimmed);
+            }
+        }
+        return keywords;
+    }
+
+    // ==================== 第三级：语义匹配 ====================
+
+    /**
+     * 语义匹配：计算问题向量，在 faq_embedding 表中做余弦距离查询
+     */
+    private Optional<FaqMatchResult> semanticMatch(String question) {
+        try {
+            EmbeddingModel embeddingModel = embeddingModelFactory.getEmbeddingModel();
+            float[] embedding = embeddingModel.call(new EmbeddingRequest(List.of(question), null))
+                    .getResult().getOutput();
+
+            // 将向量转为 PGVector 格式字符串
+            String vectorStr = toPgVectorFormat(embedding);
+
+            // 余弦距离查询：<=> 运算符返回余弦距离，相似度 = 1 - distance
+            List<Map<String, Object>> results = jdbcTemplate.queryForList(
+                    "SELECT fe.faq_id, fe.embedding <=> ?::vector AS distance, " +
+                            "kf.id, kf.question, kf.answer, kf.similar_questions, kf.category, " +
+                            "kf.status, kf.priority, kf.hit_count, kf.source, kf.create_time, kf.update_time, kf.is_delete " +
+                            "FROM faq_embedding fe " +
+                            "JOIN knowledge_faq kf ON fe.faq_id = kf.id " +
+                            "WHERE kf.status = 'ENABLED' AND kf.is_delete = false " +
+                            "ORDER BY distance ASC LIMIT 5",
+                    vectorStr
+            );
+
+            if (!results.isEmpty()) {
+                Map<String, Object> topResult = results.get(0);
+                double distance = ((Number) topResult.get("distance")).doubleValue();
+                double similarity = 1.0 - distance;
+
+                if (similarity >= semanticThreshold) {
+                    KnowledgeFaq faq = mapResultToFaq(topResult);
+                    incrementHitCount(faq.getId());
+                    return Optional.of(new FaqMatchResult(faq, "SEMANTIC", similarity));
+                }
+            }
+        } catch (Exception e) {
+            log.error("FAQ 语义匹配异常", e);
+        }
+        return Optional.empty();
+    }
+
+    // ==================== 向量化方法 ====================
+
+    /**
+     * 计算并保存单条 FAQ 的向量嵌入
+     *
+     * @param faqId    FAQ ID
+     * @param question FAQ 问题文本
+     */
+    public void computeAndSaveEmbedding(Long faqId, String question) {
+        try {
+            EmbeddingModel embeddingModel = embeddingModelFactory.getEmbeddingModel();
+            float[] embedding = embeddingModel.call(new EmbeddingRequest(List.of(question), null))
+                    .getResult().getOutput();
+
+            String vectorStr = toPgVectorFormat(embedding);
+
+            // 获取当前模型名称用于记录
+            String modelName = "unknown";
+            try {
+                var config = embeddingModel.getClass().getSimpleName();
+                modelName = config;
+            } catch (Exception ignored) {
+            }
+
+            // 先删除旧记录，再插入新记录
+            jdbcTemplate.update("DELETE FROM faq_embedding WHERE faq_id = ?", faqId);
+            jdbcTemplate.update(
+                    "INSERT INTO faq_embedding (faq_id, embedding, model_name) VALUES (?, ?::vector, ?)",
+                    faqId, vectorStr, modelName
+            );
+
+            log.info("FAQ 向量已保存: faqId={}, dimension={}", faqId, embedding.length);
+        } catch (Exception e) {
+            log.error("FAQ 向量计算/保存失败: faqId={}", faqId, e);
+        }
+    }
+
+    /**
+     * 批量异步计算 FAQ 向量嵌入
+     *
+     * @param faqIds FAQ ID 列表
+     */
+    public void batchComputeEmbeddings(List<Long> faqIds) {
+        CompletableFuture.runAsync(() -> {
+            log.info("开始批量计算 FAQ 向量: count={}", faqIds.size());
+            int success = 0;
+            int fail = 0;
+            for (Long faqId : faqIds) {
+                try {
+                    // 查询 FAQ 问题文本
+                    String question = jdbcTemplate.queryForObject(
+                            "SELECT question FROM knowledge_faq WHERE id = ? AND is_delete = false",
+                            String.class, faqId
+                    );
+                    if (question != null) {
+                        computeAndSaveEmbedding(faqId, question);
+                        success++;
+                    }
+                } catch (Exception e) {
+                    log.error("批量计算 FAQ 向量失败: faqId={}", faqId, e);
+                    fail++;
+                }
+            }
+            log.info("批量计算 FAQ 向量完成: success={}, fail={}", success, fail);
+        });
+    }
+
+    // ==================== 工具方法 ====================
+
+    /**
+     * 将 float[] 转为 PGVector 格式字符串: [0.1,0.2,0.3]
+     */
+    private String toPgVectorFormat(float[] embedding) {
+        StringBuilder sb = new StringBuilder("[");
+        for (int i = 0; i < embedding.length; i++) {
+            if (i > 0) {
+                sb.append(",");
+            }
+            sb.append(embedding[i]);
+        }
+        sb.append("]");
+        return sb.toString();
+    }
+
+    /**
+     * 增加 FAQ 命中次数
+     */
+    private void incrementHitCount(Long faqId) {
+        try {
+            jdbcTemplate.update("UPDATE knowledge_faq SET hit_count = hit_count + 1 WHERE id = ?", faqId);
+        } catch (Exception e) {
+            log.warn("更新 FAQ 命中次数失败: faqId={}", faqId, e);
+        }
+    }
+
+    /**
+     * 从 ResultSet 映射为 KnowledgeFaq 实体
+     */
+    private KnowledgeFaq mapRowToFaq(java.sql.ResultSet rs) throws java.sql.SQLException {
+        KnowledgeFaq faq = new KnowledgeFaq();
+        faq.setId(rs.getLong("id"));
+        faq.setQuestion(rs.getString("question"));
+        faq.setAnswer(rs.getString("answer"));
+        faq.setSimilarQuestions(rs.getString("similar_questions"));
+        faq.setCategory(rs.getString("category"));
+        faq.setStatus(rs.getString("status"));
+        faq.setPriority(rs.getInt("priority"));
+        faq.setHitCount(rs.getLong("hit_count"));
+        faq.setSource(rs.getString("source"));
+        faq.setCreateTime(rs.getTimestamp("create_time"));
+        faq.setUpdateTime(rs.getTimestamp("update_time"));
+        faq.setDelete(rs.getBoolean("is_delete"));
+        return faq;
+    }
+
+    /**
+     * 从 Map 结果映射为 KnowledgeFaq 实体（语义匹配使用）
+     */
+    private KnowledgeFaq mapResultToFaq(Map<String, Object> result) {
+        KnowledgeFaq faq = new KnowledgeFaq();
+        faq.setId(((Number) result.get("id")).longValue());
+        faq.setQuestion((String) result.get("question"));
+        faq.setAnswer((String) result.get("answer"));
+        faq.setSimilarQuestions((String) result.get("similar_questions"));
+        faq.setCategory((String) result.get("category"));
+        faq.setStatus((String) result.get("status"));
+        faq.setPriority(((Number) result.get("priority")).intValue());
+        faq.setHitCount(((Number) result.get("hit_count")).longValue());
+        faq.setSource((String) result.get("source"));
+        faq.setCreateTime(result.get("create_time") instanceof java.sql.Timestamp ts ? new Date(ts.getTime()) : null);
+        faq.setUpdateTime(result.get("update_time") instanceof java.sql.Timestamp ts ? new Date(ts.getTime()) : null);
+        faq.setDelete(Boolean.TRUE.equals(result.get("is_delete")));
+        return faq;
+    }
+}

src/main/java/com/wok/supportbot/service/FaqService.java

@@ -0,0 +1,299 @@
+package com.wok.supportbot.service;
+
+import com.baomidou.mybatisplus.core.conditions.query.QueryWrapper;
+import com.fasterxml.jackson.core.type.TypeReference;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.wok.supportbot.dao.KnowledgeFaqMapper;
+import com.wok.supportbot.entity.KnowledgeFaq;
+import lombok.extern.slf4j.Slf4j;
+import org.springframework.beans.factory.annotation.Autowired;
+import org.springframework.jdbc.core.JdbcTemplate;
+import org.springframework.stereotype.Service;
+import org.springframework.transaction.annotation.Transactional;
+
+import java.util.*;
+import java.util.concurrent.CompletableFuture;
+
+/**
+ * FAQ 知识库 CRUD 服务
+ */
+@Service
+@Slf4j
+public class FaqService {
+
+    @Autowired
+    private KnowledgeFaqMapper faqMapper;
+
+    @Autowired
+    private JdbcTemplate jdbcTemplate;
+
+    @Autowired
+    private FaqMatchEngine faqMatchEngine;
+
+    private final ObjectMapper objectMapper = new ObjectMapper();
+
+    // ==================== 分页查询 ====================
+
+    /**
+     * 分页查询 FAQ 列表
+     *
+     * @param page     页码（从1开始）
+     * @param size     每页条数
+     * @param keyword  关键词搜索（question LIKE，可选）
+     * @param category 分类过滤（可选）
+     * @param status   状态过滤（可选）
+     * @return 分页结果 Map {total, records}
+     */
+    public Map<String, Object> list(int page, int size, String keyword, String category, String status) {
+        QueryWrapper<KnowledgeFaq> wrapper = new QueryWrapper<>();
+
+        if (keyword != null && !keyword.isBlank()) {
+            wrapper.like("question", keyword.trim());
+        }
+        if (category != null && !category.isBlank()) {
+            wrapper.eq("category", category.trim());
+        }
+        if (status != null && !status.isBlank()) {
+            wrapper.eq("status", status.trim());
+        }
+
+        Long total = faqMapper.selectCount(wrapper);
+
+        wrapper.orderByDesc("priority").orderByDesc("create_time");
+        wrapper.last("LIMIT " + size + " OFFSET " + (page - 1) * size);
+        List<KnowledgeFaq> records = faqMapper.selectList(wrapper);
+
+        Map<String, Object> result = new HashMap<>();
+        result.put("total", total);
+        result.put("records", records);
+        return result;
+    }
+
+    // ==================== 新增 ====================
+
+    /**
+     * 新增 FAQ，新增后异步计算向量
+     */
+    @Transactional(rollbackFor = Exception.class)
+    public KnowledgeFaq create(KnowledgeFaq faq) {
+        // 设置默认值
+        if (faq.getStatus() == null || faq.getStatus().isBlank()) {
+            faq.setStatus("ENABLED");
+        }
+        if (faq.getPriority() == null) {
+            faq.setPriority(0);
+        }
+        if (faq.getHitCount() == null) {
+            faq.setHitCount(0L);
+        }
+        if (faq.getSource() == null || faq.getSource().isBlank()) {
+            faq.setSource("manual");
+        }
+        if (faq.getSimilarQuestions() == null) {
+            faq.setSimilarQuestions("[]");
+        }
+        faq.setCreateTime(new Date());
+        faq.setUpdateTime(new Date());
+
+        faqMapper.insert(faq);
+        log.info("FAQ 已创建: id={}, question={}", faq.getId(), faq.getQuestion());
+
+        // 异步计算向量
+        CompletableFuture.runAsync(() -> faqMatchEngine.computeAndSaveEmbedding(faq.getId(), faq.getQuestion()));
+
+        return faq;
+    }
+
+    // ==================== 修改 ====================
+
+    /**
+     * 修改 FAQ，修改后重新计算向量
+     */
+    @Transactional(rollbackFor = Exception.class)
+    public KnowledgeFaq update(Long id, KnowledgeFaq faq) {
+        KnowledgeFaq existing = faqMapper.selectById(id);
+        if (existing == null) {
+            throw new IllegalArgumentException("FAQ 不存在: id=" + id);
+        }
+
+        // 更新非空字段
+        if (faq.getQuestion() != null) {
+            existing.setQuestion(faq.getQuestion());
+        }
+        if (faq.getAnswer() != null) {
+            existing.setAnswer(faq.getAnswer());
+        }
+        if (faq.getSimilarQuestions() != null) {
+            existing.setSimilarQuestions(faq.getSimilarQuestions());
+        }
+        if (faq.getCategory() != null) {
+            existing.setCategory(faq.getCategory());
+        }
+        if (faq.getStatus() != null) {
+            existing.setStatus(faq.getStatus());
+        }
+        if (faq.getPriority() != null) {
+            existing.setPriority(faq.getPriority());
+        }
+        existing.setUpdateTime(new Date());
+
+        faqMapper.updateById(existing);
+        log.info("FAQ 已更新: id={}", id);
+
+        // 问题文本变更时重新计算向量
+        CompletableFuture.runAsync(() -> faqMatchEngine.computeAndSaveEmbedding(id, existing.getQuestion()));
+
+        return existing;
+    }
+
+    // ==================== 删除 ====================
+
+    /**
+     * 逻辑删除 FAQ，同时删除 faq_embedding 中对应记录
+     */
+    @Transactional(rollbackFor = Exception.class)
+    public void delete(Long id) {
+        KnowledgeFaq existing = faqMapper.selectById(id);
+        if (existing == null) {
+            throw new IllegalArgumentException("FAQ 不存在: id=" + id);
+        }
+
+        // 逻辑删除 FAQ
+        faqMapper.deleteById(id);
+
+        // 物理删除对应的向量记录
+        jdbcTemplate.update("DELETE FROM faq_embedding WHERE faq_id = ?", id);
+
+        log.info("FAQ 已删除: id={}", id);
+    }
+
+    // ==================== 启用/禁用 ====================
+
+    /**
+     * 切换 FAQ 启用/禁用状态
+     */
+    @Transactional(rollbackFor = Exception.class)
+    public void toggleStatus(Long id, String status) {
+        KnowledgeFaq existing = faqMapper.selectById(id);
+        if (existing == null) {
+            throw new IllegalArgumentException("FAQ 不存在: id=" + id);
+        }
+        if (!"ENABLED".equals(status) && !"DISABLED".equals(status)) {
+            throw new IllegalArgumentException("无效的状态值: " + status + "，仅支持 ENABLED/DISABLED");
+        }
+
+        existing.setStatus(status);
+        existing.setUpdateTime(new Date());
+        faqMapper.updateById(existing);
+
+        log.info("FAQ 状态已切换: id={}, status={}", id, status);
+    }
+
+    // ==================== 批量导入 ====================
+
+    /**
+     * 批量导入 FAQ（Excel 解析后的数据），批量计算向量
+     */
+    @Transactional(rollbackFor = Exception.class)
+    public int batchImport(List<KnowledgeFaq> faqs) {
+        if (faqs == null || faqs.isEmpty()) {
+            return 0;
+        }
+
+        List<Long> importedIds = new ArrayList<>();
+        for (KnowledgeFaq faq : faqs) {
+            // 设置默认值
+            if (faq.getStatus() == null || faq.getStatus().isBlank()) {
+                faq.setStatus("ENABLED");
+            }
+            if (faq.getPriority() == null) {
+                faq.setPriority(0);
+            }
+            if (faq.getHitCount() == null) {
+                faq.setHitCount(0L);
+            }
+            if (faq.getSource() == null || faq.getSource().isBlank()) {
+                faq.setSource("import");
+            }
+            if (faq.getSimilarQuestions() == null) {
+                faq.setSimilarQuestions("[]");
+            }
+            faq.setCreateTime(new Date());
+            faq.setUpdateTime(new Date());
+
+            faqMapper.insert(faq);
+            importedIds.add(faq.getId());
+        }
+
+        log.info("FAQ 批量导入完成: count={}", importedIds.size());
+
+        // 异步批量计算向量
+        faqMatchEngine.batchComputeEmbeddings(importedIds);
+
+        return importedIds.size();
+    }
+
+    // ==================== 导出 ====================
+
+    /**
+     * 导出所有启用的 FAQ
+     */
+    public List<KnowledgeFaq> exportAll() {
+        QueryWrapper<KnowledgeFaq> wrapper = new QueryWrapper<>();
+        wrapper.eq("status", "ENABLED");
+        wrapper.orderByDesc("priority").orderByDesc("hit_count");
+        return faqMapper.selectList(wrapper);
+    }
+
+    // ==================== 统计 ====================
+
+    /**
+     * 获取 FAQ 匹配统计信息
+     *
+     * @return 统计数据 Map
+     */
+    public Map<String, Object> getStats() {
+        Map<String, Object> stats = new HashMap<>();
+
+        // 总 FAQ 数
+        Long totalCount = faqMapper.selectCount(new QueryWrapper<>());
+        stats.put("totalCount", totalCount);
+
+        // 启用数
+        QueryWrapper<KnowledgeFaq> enabledWrapper = new QueryWrapper<>();
+        enabledWrapper.eq("status", "ENABLED");
+        Long enabledCount = faqMapper.selectCount(enabledWrapper);
+        stats.put("enabledCount", enabledCount);
+
+        // 总命中次数
+        Long totalHits = jdbcTemplate.queryForObject(
+                "SELECT COALESCE(SUM(hit_count), 0) FROM knowledge_faq WHERE is_delete = false",
+                Long.class
+        );
+        stats.put("totalHits", totalHits);
+
+        // Top10 热门 FAQ
+        List<Map<String, Object>> topFaqs = jdbcTemplate.queryForList(
+                "SELECT id, question, hit_count, category FROM knowledge_faq " +
+                        "WHERE is_delete = false AND status = 'ENABLED' " +
+                        "ORDER BY hit_count DESC LIMIT 10"
+        );
+        stats.put("topFaqs", topFaqs);
+
+        return stats;
+    }
+
+    // ==================== 手动重算向量 ====================
+
+    /**
+     * 手动重新计算某条 FAQ 的向量
+     */
+    public void recomputeEmbedding(Long id) {
+        KnowledgeFaq existing = faqMapper.selectById(id);
+        if (existing == null) {
+            throw new IllegalArgumentException("FAQ 不存在: id=" + id);
+        }
+        faqMatchEngine.computeAndSaveEmbedding(id, existing.getQuestion());
+        log.info("FAQ 向量已重新计算: id={}", id);
+    }
+}

src/main/java/com/wok/supportbot/service/IntentRouter.java

@@ -0,0 +1,119 @@
+package com.wok.supportbot.service;
+
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.wok.supportbot.config.ChatModelFactory;
+import lombok.AllArgsConstructor;
+import lombok.Data;
+import lombok.NoArgsConstructor;
+import lombok.extern.slf4j.Slf4j;
+import org.springframework.ai.chat.model.ChatModel;
+import org.springframework.ai.chat.prompt.Prompt;
+import org.springframework.beans.factory.annotation.Autowired;
+import org.springframework.stereotype.Service;
+
+/**
+ * LLM 意图分类路由器
+ * 使用 ChatModel 对用户问题进行意图分类，决定后续处理流程：
+ * - FAQ: 常见问题 → FaqMatchEngine 精准匹配
+ * - RAG: 知识库检索 → 现有 RAG 流程
+ * - CHITCHAT: 闲聊 → 简单对话
+ */
+@Service
+@Slf4j
+public class IntentRouter {
+
+    @Autowired
+    private ChatModelFactory chatModelFactory;
+
+    private final ObjectMapper objectMapper = new ObjectMapper();
+
+    /** 意图分类 Prompt 模板 */
+    private static final String INTENT_PROMPT_TEMPLATE = """
+            你是一个意图分类器。根据用户问题，判断其属于以下哪个意图：
+            - FAQ: 常见问题，如产品功能、价格、退换货政策、服务流程等标准问答
+            - RAG: 需要查阅文档/知识库才能回答的专业问题或细节问题
+            - CHITCHAT: 闲聊、问候、感谢、告别等非业务话题
+
+            仅返回JSON格式: {"intent":"FAQ|RAG|CHITCHAT","confidence":0.0-1.0}
+            用户问题: %s
+            """;
+
+    // ==================== 意图结果内部类 ====================
+
+    /**
+     * 意图分类结果
+     */
+    @Data
+    @AllArgsConstructor
+    @NoArgsConstructor
+    public static class IntentResult {
+        /** 意图类型: FAQ / RAG / CHITCHAT */
+        private String intent;
+        /** 置信度 (0.0 ~ 1.0) */
+        private double confidence;
+    }
+
+    // ==================== 核心路由方法 ====================
+
+    /**
+     * 对用户问题进行意图分类
+     *
+     * @param userQuestion 用户问题
+     * @return 意图分类结果
+     */
+    public IntentResult route(String userQuestion) {
+        if (userQuestion == null || userQuestion.isBlank()) {
+            return new IntentResult("RAG", 0.0);
+        }
+
+        try {
+            ChatModel chatModel = chatModelFactory.getChatModel("CHAT");
+            String promptText = INTENT_PROMPT_TEMPLATE.formatted(userQuestion);
+            Prompt prompt = new Prompt(promptText);
+
+            String response = chatModel.call(prompt).getResult().getOutput().getText();
+            log.debug("意图分类原始响应: {}", response);
+
+            return parseIntentResponse(response);
+        } catch (Exception e) {
+            log.error("意图分类失败，降级为 RAG: question={}", userQuestion, e);
+            return new IntentResult("RAG", 0.0);
+        }
+    }
+
+    // ==================== 解析方法 ====================
+
+    /**
+     * 解析 LLM 返回的 JSON 意图分类结果
+     * 如果解析失败，默认返回 RAG（降级到现有流程）
+     */
+    private IntentResult parseIntentResponse(String response) {
+        try {
+            // 清理可能的 markdown 代码块包裹
+            String cleaned = response.trim();
+            if (cleaned.startsWith("```")) {
+                cleaned = cleaned.replaceAll("^```(?:json)?\\s*", "").replaceAll("\\s*```$", "");
+            }
+
+            IntentResult result = objectMapper.readValue(cleaned, IntentResult.class);
+
+            // 校验意图类型有效性
+            if (result.getIntent() == null || !isValidIntent(result.getIntent())) {
+                log.warn("无效的意图类型: {}, 降级为 RAG", result.getIntent());
+                return new IntentResult("RAG", 0.5);
+            }
+
+            return result;
+        } catch (Exception e) {
+            log.warn("意图分类 JSON 解析失败，降级为 RAG: response={}", response, e);
+            return new IntentResult("RAG", 0.0);
+        }
+    }
+
+    /**
+     * 校验意图类型是否有效
+     */
+    private boolean isValidIntent(String intent) {
+        return "FAQ".equals(intent) || "RAG".equals(intent) || "CHITCHAT".equals(intent);
+    }
+}

src/main/java/com/wok/supportbot/service/MessageFeedbackService.java

@@ -0,0 +1,178 @@
+package com.wok.supportbot.service;
+
+import com.baomidou.mybatisplus.core.conditions.query.LambdaQueryWrapper;
+import com.wok.supportbot.dao.MessageFeedbackMapper;
+import com.wok.supportbot.entity.MessageFeedback;
+import lombok.extern.slf4j.Slf4j;
+import org.springframework.beans.factory.annotation.Autowired;
+import org.springframework.jdbc.core.JdbcTemplate;
+import org.springframework.stereotype.Service;
+
+import java.util.*;
+
+/**
+ * 消息反馈服务
+ * 提供反馈提交、查询、统计等功能
+ */
+@Service
+@Slf4j
+public class MessageFeedbackService {
+
+    @Autowired
+    private MessageFeedbackMapper messageFeedbackMapper;
+
+    @Autowired
+    private JdbcTemplate jdbcTemplate;
+
+    /**
+     * 提交/修改反馈（upsert 语义）
+     * 按 messageId 查询，存在则更新（覆盖上次），不存在则插入
+     *
+     * @param feedback 反馈信息
+     * @return 保存后的反馈实体
+     */
+    public MessageFeedback submitFeedback(MessageFeedback feedback) {
+        // 按 messageId 查询已有反馈
+        LambdaQueryWrapper<MessageFeedback> wrapper = new LambdaQueryWrapper<>();
+        wrapper.eq(MessageFeedback::getMessageId, feedback.getMessageId());
+        MessageFeedback existing = messageFeedbackMapper.selectOne(wrapper);
+
+        if (existing != null) {
+            // 更新已有反馈
+            existing.setFeedbackType(feedback.getFeedbackType());
+            existing.setReasonCategory(feedback.getReasonCategory());
+            existing.setReasonComment(feedback.getReasonComment());
+            existing.setUpdateTime(new Date());
+            messageFeedbackMapper.updateById(existing);
+            log.info("更新反馈: messageId={}, type={}", feedback.getMessageId(), feedback.getFeedbackType());
+            return existing;
+        } else {
+            // 新增反馈
+            feedback.setCreateTime(new Date());
+            feedback.setUpdateTime(new Date());
+            messageFeedbackMapper.insert(feedback);
+            log.info("新增反馈: messageId={}, type={}", feedback.getMessageId(), feedback.getFeedbackType());
+            return feedback;
+        }
+    }
+
+    /**
+     * 按会话ID查询所有反馈
+     *
+     * @param conversationId 会话ID
+     * @return 反馈列表
+     */
+    public List<MessageFeedback> getByConversationId(String conversationId) {
+        LambdaQueryWrapper<MessageFeedback> wrapper = new LambdaQueryWrapper<>();
+        wrapper.eq(MessageFeedback::getConversationId, conversationId)
+                .orderByDesc(MessageFeedback::getCreateTime);
+        return messageFeedbackMapper.selectList(wrapper);
+    }
+
+    /**
+     * 按消息ID查询反馈
+     *
+     * @param messageId 消息ID
+     * @return 反馈实体（可能为 null）
+     */
+    public MessageFeedback getByMessageId(String messageId) {
+        LambdaQueryWrapper<MessageFeedback> wrapper = new LambdaQueryWrapper<>();
+        wrapper.eq(MessageFeedback::getMessageId, messageId);
+        return messageFeedbackMapper.selectOne(wrapper);
+    }
+
+    /**
+     * 批量查询反馈状态（供 SDK 回显使用）
+     *
+     * @param messageIds 消息ID列表
+     * @return 反馈列表
+     */
+    public List<MessageFeedback> getBatchByMessageIds(List<String> messageIds) {
+        if (messageIds == null || messageIds.isEmpty()) {
+            return Collections.emptyList();
+        }
+        LambdaQueryWrapper<MessageFeedback> wrapper = new LambdaQueryWrapper<>();
+        wrapper.in(MessageFeedback::getMessageId, messageIds);
+        return messageFeedbackMapper.selectList(wrapper);
+    }
+
+    /**
+     * 获取反馈统计数据
+     *
+     * @param startDate 开始日期（yyyy-MM-dd），可选
+     * @param endDate   结束日期（yyyy-MM-dd），可选
+     * @return 统计结果 Map
+     */
+    public Map<String, Object> getStats(String startDate, String endDate) {
+        Map<String, Object> result = new LinkedHashMap<>();
+
+        // 构建日期过滤条件
+        StringBuilder whereClause = new StringBuilder("WHERE is_delete = false ");
+        List<Object> params = new ArrayList<>();
+        if (startDate != null && !startDate.isBlank()) {
+            whereClause.append("AND create_time >= ?::timestamp ");
+            params.add(startDate + " 00:00:00");
+        }
+        if (endDate != null && !endDate.isBlank()) {
+            whereClause.append("AND create_time <= ?::timestamp ");
+            params.add(endDate + " 23:59:59");
+        }
+
+        // 总反馈数、👍数、👎数
+        String countSql = """
+                SELECT
+                    COUNT(*) AS total,
+                    COUNT(*) FILTER (WHERE feedback_type = 'THUMBS_UP') AS thumbs_up,
+                    COUNT(*) FILTER (WHERE feedback_type = 'THUMBS_DOWN') AS thumbs_down
+                FROM message_feedback
+                """ + whereClause;
+        Map<String, Object> counts = jdbcTemplate.queryForMap(countSql, params.toArray());
+        long totalFeedbacks = ((Number) counts.get("total")).longValue();
+        long thumbsUpCount = ((Number) counts.get("thumbs_up")).longValue();
+        long thumbsDownCount = ((Number) counts.get("thumbs_down")).longValue();
+
+        result.put("totalFeedbacks", totalFeedbacks);
+        result.put("thumbsUpCount", thumbsUpCount);
+        result.put("thumbsDownCount", thumbsDownCount);
+        result.put("satisfactionRate", totalFeedbacks > 0
+                ? Math.round((double) thumbsUpCount / totalFeedbacks * 100.0) / 100.0
+                : 0.0);
+
+        // 各原因分布（仅统计 THUMBS_DOWN）
+        String reasonSql = """
+                SELECT reason_category, COUNT(*) AS cnt
+                FROM message_feedback
+                """ + whereClause + " AND feedback_type = 'THUMBS_DOWN' AND reason_category IS NOT NULL " +
+                "GROUP BY reason_category ORDER BY cnt DESC";
+        List<Map<String, Object>> reasonRows = jdbcTemplate.queryForList(reasonSql, params.toArray());
+        Map<String, Long> reasonDistribution = new LinkedHashMap<>();
+        for (Map<String, Object> row : reasonRows) {
+            String category = (String) row.get("reason_category");
+            long cnt = ((Number) row.get("cnt")).longValue();
+            reasonDistribution.put(category, cnt);
+        }
+        result.put("reasonDistribution", reasonDistribution);
+
+        // 按天趋势
+        String trendSql = """
+                SELECT
+                    DATE(create_time) AS date,
+                    COUNT(*) FILTER (WHERE feedback_type = 'THUMBS_UP') AS up,
+                    COUNT(*) FILTER (WHERE feedback_type = 'THUMBS_DOWN') AS down
+                FROM message_feedback
+                """ + whereClause +
+                " GROUP BY DATE(create_time) ORDER BY date ASC";
+        List<Map<String, Object>> trendRows = jdbcTemplate.queryForList(trendSql, params.toArray());
+        List<Map<String, Object>> dailyTrends = new ArrayList<>();
+        for (Map<String, Object> row : trendRows) {
+            Map<String, Object> trend = new LinkedHashMap<>();
+            trend.put("date", String.valueOf(row.get("date")));
+            trend.put("up", ((Number) row.get("up")).longValue());
+            trend.put("down", ((Number) row.get("down")).longValue());
+            dailyTrends.add(trend);
+        }
+        result.put("dailyTrends", dailyTrends);
+
+        return result;
+    }
+}

src/main/java/com/wok/supportbot/service/SensitiveWordService.java

@@ -0,0 +1,209 @@
+package com.wok.supportbot.service;
+
+import com.baomidou.mybatisplus.core.conditions.query.QueryWrapper;
+import com.wok.supportbot.dao.SensitiveWordMapper;
+import com.wok.supportbot.entity.SensitiveWord;
+import lombok.extern.slf4j.Slf4j;
+import org.springframework.beans.factory.annotation.Autowired;
+import org.springframework.stereotype.Service;
+
+import java.util.ArrayList;
+import java.util.HashMap;
+import java.util.List;
+import java.util.Map;
+
+/**
+ * 敏感词 CRUD 服务
+ */
+@Slf4j
+@Service
+public class SensitiveWordService {
+
+    @Autowired
+    private SensitiveWordMapper sensitiveWordMapper;
+
+    @Autowired
+    private ContentSafetyService contentSafetyService;
+
+    /**
+     * 分页查询敏感词列表
+     *
+     * @param page     页码（从1开始）
+     * @param size     每页条数
+     * @param keyword  关键词搜索（匹配 word 字段）
+     * @param category 分类筛选
+     * @return 包含 records/total/page/size/pages 的结果
+     */
+    public Map<String, Object> list(int page, int size, String keyword, String category) {
+        // 构建计数条件
+        QueryWrapper<SensitiveWord> countWrapper = new QueryWrapper<>();
+        if (keyword != null && !keyword.isBlank()) {
+            countWrapper.like("word", keyword);
+        }
+        if (category != null && !category.isBlank()) {
+            countWrapper.eq("category", category);
+        }
+        Long total = sensitiveWordMapper.selectCount(countWrapper);
+
+        // 构建列表查询条件
+        QueryWrapper<SensitiveWord> listWrapper = new QueryWrapper<>();
+        if (keyword != null && !keyword.isBlank()) {
+            listWrapper.like("word", keyword);
+        }
+        if (category != null && !category.isBlank()) {
+            listWrapper.eq("category", category);
+        }
+        listWrapper.orderByDesc("create_time");
+        listWrapper.last("LIMIT " + size + " OFFSET " + (long) (page - 1) * size);
+        List<SensitiveWord> records = sensitiveWordMapper.selectList(listWrapper);
+
+        // 组装返回结果
+        Map<String, Object> result = new HashMap<>();
+        result.put("records", records);
+        result.put("total", total);
+        result.put("page", page);
+        result.put("size", size);
+        result.put("pages", (total + size - 1) / size);
+        return result;
+    }
+
+    /**
+     * 新增敏感词（检查重复）
+     *
+     * @param word 敏感词实体
+     * @return 新增后的实体
+     */
+    public SensitiveWord create(SensitiveWord word) {
+        // 检查是否已存在相同的敏感词
+        QueryWrapper<SensitiveWord> wrapper = new QueryWrapper<>();
+        wrapper.eq("word", word.getWord());
+        Long count = sensitiveWordMapper.selectCount(wrapper);
+        if (count > 0) {
+            throw new IllegalArgumentException("敏感词已存在：" + word.getWord());
+        }
+
+        // 设置默认值
+        if (word.getCategory() == null || word.getCategory().isBlank()) {
+            word.setCategory("custom");
+        }
+        if (word.getLevel() == null) {
+            word.setLevel(1);
+        }
+        if (word.getIsActive() == null) {
+            word.setIsActive(true);
+        }
+
+        sensitiveWordMapper.insert(word);
+
+        // 刷新 DFA 字典树
+        contentSafetyService.rebuild();
+        log.info("新增敏感词：{}", word.getWord());
+        return word;
+    }
+
+    /**
+     * 修改敏感词
+     *
+     * @param id   敏感词ID
+     * @param word 更新内容
+     * @return 更新后的实体
+     */
+    public SensitiveWord update(Long id, SensitiveWord word) {
+        SensitiveWord existing = sensitiveWordMapper.selectById(id);
+        if (existing == null) {
+            throw new IllegalArgumentException("敏感词不存在，ID：" + id);
+        }
+
+        // 如果修改了 word 内容，检查新词是否与其他记录重复
+        if (word.getWord() != null && !word.getWord().equals(existing.getWord())) {
+            QueryWrapper<SensitiveWord> wrapper = new QueryWrapper<>();
+            wrapper.eq("word", word.getWord());
+            wrapper.ne("id", id);
+            Long count = sensitiveWordMapper.selectCount(wrapper);
+            if (count > 0) {
+                throw new IllegalArgumentException("敏感词已存在：" + word.getWord());
+            }
+        }
+
+        word.setId(id);
+        sensitiveWordMapper.updateById(word);
+
+        // 刷新 DFA 字典树
+        contentSafetyService.rebuild();
+        log.info("修改敏感词，ID：{}", id);
+        return sensitiveWordMapper.selectById(id);
+    }
+
+    /**
+     * 逻辑删除敏感词
+     *
+     * @param id 敏感词ID
+     */
+    public void delete(Long id) {
+        SensitiveWord existing = sensitiveWordMapper.selectById(id);
+        if (existing == null) {
+            throw new IllegalArgumentException("敏感词不存在，ID：" + id);
+        }
+
+        sensitiveWordMapper.deleteById(id);
+
+        // 刷新 DFA 字典树
+        contentSafetyService.rebuild();
+        log.info("删除敏感词，ID：{}，词：{}", id, existing.getWord());
+    }
+
+    /**
+     * 批量导入敏感词
+     *
+     * @param words    敏感词列表
+     * @param category 分类
+     * @param level    级别
+     * @return 成功导入的数量
+     */
+    public int batchImport(List<String> words, String category, int level) {
+        if (words == null || words.isEmpty()) {
+            return 0;
+        }
+
+        String cat = (category != null && !category.isBlank()) ? category : "custom";
+        int imported = 0;
+        List<SensitiveWord> toInsert = new ArrayList<>();
+
+        for (String w : words) {
+            String trimmed = w.trim();
+            if (trimmed.isEmpty()) {
+                continue;
+            }
+
+            // 检查是否已存在
+            QueryWrapper<SensitiveWord> wrapper = new QueryWrapper<>();
+            wrapper.eq("word", trimmed);
+            Long count = sensitiveWordMapper.selectCount(wrapper);
+            if (count > 0) {
+                continue; // 跳过重复词
+            }
+
+            SensitiveWord sw = SensitiveWord.builder()
+                    .word(trimmed)
+                    .category(cat)
+                    .level(level)
+                    .isActive(true)
+                    .build();
+            toInsert.add(sw);
+            imported++;
+        }
+
+        // 批量插入
+        for (SensitiveWord sw : toInsert) {
+            sensitiveWordMapper.insert(sw);
+        }
+
+        // 刷新 DFA 字典树
+        if (imported > 0) {
+            contentSafetyService.rebuild();
+            log.info("批量导入敏感词完成，成功 {} 个，共提交 {} 个", imported, words.size());
+        }
+
+        return imported;
+    }
+}

src/main/resources/add-comments.sql

@@ -1,3 +1,9 @@
+-- ================================================================
+-- 旧版注释补丁脚本（已合并到 init-database.sql）
+-- 如需为已存在的表补充注释，请使用桌面上的 add-table-comments.sql
+-- 或直接参考 init-database.sql 中的 COMMENT ON 语句
+-- ================================================================
+
 -- 重新设置 chat_message 表注释
 COMMENT ON TABLE chat_message IS '聊天消息表（存储用户与AI助手的对话历史）';
 

src/main/resources/application.yml

@@ -61,6 +61,9 @@ knowledge:
   role:
     # 严格隔离：true=角色未绑定知识库分类时禁止检索任何内容；false=可检索全部知识库
     strict-isolation: false
+  faq:
+    # FAQ 语义匹配阈值（0-1），低于此值自动降级到 RAG 检索
+    semantic-threshold: 0.85
 
 # ==================== Knife4j API 文档通用配置 ====================
 # knife4j.enable 开关在各环境 yml 中配置（生产环境建议关闭）

src/main/resources/init-database.sql

@@ -1,6 +1,7 @@
 -- ============================================================
--- AI 智能客服系统 - 数据库初始化脚本
+-- AI 智能客服系统 - 数据库初始化脚本（完整版）
 -- 适用环境: PostgreSQL 12+ 且已安装 pgvector 扩展
+-- 表数量: 14 张（基础 8 张 + P0 阶段新增 6 张）
 --
 -- 使用方法:
 --   方式一（psql 命令行）:
@@ -287,7 +288,212 @@ COMMENT ON COLUMN vector_store.embedding IS '向量嵌入（维度 1024）';
 COMMENT ON COLUMN vector_store.create_time IS '创建时间';
 COMMENT ON COLUMN vector_store.update_time IS '更新时间';
 
+-- 全文检索列（混合检索 KEYWORD / HYBRID 模式使用）
+ALTER TABLE vector_store ADD COLUMN IF NOT EXISTS content_tsvector tsvector;
+CREATE INDEX IF NOT EXISTS idx_vector_store_tsvector ON vector_store USING gin(content_tsvector);
+
+COMMENT ON COLUMN vector_store.content_tsvector IS '全文检索向量列（tsvector，由触发器自动维护）';
+
+-- 全文检索触发器函数
+CREATE OR REPLACE FUNCTION update_content_tsvector() RETURNS trigger AS $$
+BEGIN
+    NEW.content_tsvector := to_tsvector('simple', coalesce(NEW.content, ''));
+    RETURN NEW;
+END;
+$$ LANGUAGE plpgsql;
+
+DROP TRIGGER IF EXISTS trg_update_content_tsvector ON vector_store;
+CREATE TRIGGER trg_update_content_tsvector
+    BEFORE INSERT OR UPDATE ON vector_store
+    FOR EACH ROW EXECUTE FUNCTION update_content_tsvector();
+
+-- ============================================================
+-- 表 9: ai_model_config — AI 大模型配置表
+-- ============================================================
+CREATE TABLE IF NOT EXISTS ai_model_config (
+    id              BIGINT       PRIMARY KEY,
+    name            VARCHAR(100) NOT NULL,
+    app_type        VARCHAR(32)  NOT NULL,
+    provider        VARCHAR(64)  NOT NULL,
+    api_key         VARCHAR(256),
+    model_name      VARCHAR(128) NOT NULL,
+    temperature     DOUBLE PRECISION DEFAULT 0.7,
+    max_tokens      INTEGER      DEFAULT 2048,
+    base_url        VARCHAR(256),
+    extra_config    JSONB        DEFAULT '{}' NOT NULL,
+    is_active       BOOLEAN      DEFAULT FALSE NOT NULL,
+    priority        INTEGER      DEFAULT 0 NOT NULL,
+    description     VARCHAR(500),
+    create_time     TIMESTAMP    NOT NULL DEFAULT CURRENT_TIMESTAMP,
+    update_time     TIMESTAMP    NOT NULL DEFAULT CURRENT_TIMESTAMP,
+    is_delete       BOOLEAN      NOT NULL DEFAULT FALSE
+);
+
+CREATE INDEX IF NOT EXISTS idx_ai_model_config_app_type  ON ai_model_config (app_type);
+CREATE INDEX IF NOT EXISTS idx_ai_model_config_active    ON ai_model_config (is_active) WHERE is_delete = FALSE;
+
+COMMENT ON TABLE  ai_model_config              IS 'AI 大模型配置表（管理多套模型配置，按应用类型绑定）';
+COMMENT ON COLUMN ai_model_config.id           IS '主键（雪花算法生成）';
+COMMENT ON COLUMN ai_model_config.name         IS '配置名称';
+COMMENT ON COLUMN ai_model_config.app_type     IS '应用类型: CHAT / EMBEDDING / RAG_REWRITE / RERANK';
+COMMENT ON COLUMN ai_model_config.provider     IS '模型提供商: dashscope / openai / deepseek / moonshot / zhipu / volcengine 等';
+COMMENT ON COLUMN ai_model_config.api_key      IS 'API Key（数据库加密存储，前端脱敏展示）';
+COMMENT ON COLUMN ai_model_config.model_name   IS '模型名称（如 qwen-turbo、gpt-4o）';
+COMMENT ON COLUMN ai_model_config.temperature  IS '温度参数（控制输出随机性，0.0~2.0）';
+COMMENT ON COLUMN ai_model_config.max_tokens   IS '最大 Token 数（单次生成上限）';
+COMMENT ON COLUMN ai_model_config.base_url     IS 'API 基础地址（可选，用于私有化部署或第三方厂商）';
+COMMENT ON COLUMN ai_model_config.extra_config IS '扩展配置（JSONB，存储 topP、dimensions 等自定义参数）';
+COMMENT ON COLUMN ai_model_config.is_active    IS '是否激活（每种 App 类型只能有一个激活配置）';
+COMMENT ON COLUMN ai_model_config.priority     IS '优先级（数值越大越优先，多套配置时生效）';
+COMMENT ON COLUMN ai_model_config.description  IS '配置描述说明';
+COMMENT ON COLUMN ai_model_config.create_time  IS '创建时间';
+COMMENT ON COLUMN ai_model_config.update_time  IS '更新时间';
+COMMENT ON COLUMN ai_model_config.is_delete    IS '逻辑删除: FALSE=正常 TRUE=已删除';
+
+-- ============================================================
+-- 表 10: sensitive_word — 敏感词表（P0-004 内容安全过滤）
+-- ============================================================
+CREATE TABLE IF NOT EXISTS sensitive_word (
+    id              BIGSERIAL    PRIMARY KEY,
+    word            VARCHAR(256) NOT NULL,
+    category        VARCHAR(64)  NOT NULL DEFAULT 'custom',
+    level           INTEGER      NOT NULL DEFAULT 1,
+    is_active       BOOLEAN      NOT NULL DEFAULT TRUE,
+    remark          VARCHAR(512),
+    create_time     TIMESTAMP    NOT NULL DEFAULT CURRENT_TIMESTAMP,
+    update_time     TIMESTAMP    NOT NULL DEFAULT CURRENT_TIMESTAMP,
+    is_delete       BOOLEAN      NOT NULL DEFAULT FALSE
+);
+
+-- 唯一索引：同一分类下敏感词不重复（仅对未删除记录生效）
+CREATE UNIQUE INDEX IF NOT EXISTS uk_sensitive_word_word_category ON sensitive_word (word, category) WHERE is_delete = FALSE;
+CREATE INDEX IF NOT EXISTS idx_sensitive_word_active ON sensitive_word (is_active) WHERE is_delete = FALSE;
+
+COMMENT ON TABLE  sensitive_word            IS '敏感词表（DFA 引擎驱动的内容安全过滤）';
+COMMENT ON COLUMN sensitive_word.id         IS '主键（雪花算法生成）';
+COMMENT ON COLUMN sensitive_word.word       IS '敏感词内容';
+COMMENT ON COLUMN sensitive_word.category   IS '分类: politics / porn / abuse / custom，默认 custom';
+COMMENT ON COLUMN sensitive_word.level      IS '级别: 1=脱敏(MASK) / 2=拦截(BLOCK)';
+COMMENT ON COLUMN sensitive_word.is_active  IS '是否启用';
+COMMENT ON COLUMN sensitive_word.remark     IS '备注说明';
+COMMENT ON COLUMN sensitive_word.create_time IS '创建时间';
+COMMENT ON COLUMN sensitive_word.update_time IS '更新时间';
+COMMENT ON COLUMN sensitive_word.is_delete  IS '逻辑删除: FALSE=正常 TRUE=已删除';
+
+-- ============================================================
+-- 表 11: content_audit_log — 内容审计日志表（P0-004，只追加不删除）
+-- ============================================================
+CREATE TABLE IF NOT EXISTS content_audit_log (
+    id              BIGSERIAL    PRIMARY KEY,
+    session_id      VARCHAR(64),
+    direction       VARCHAR(16)  NOT NULL,
+    original_text   TEXT         NOT NULL,
+    hit_words       JSONB        NOT NULL DEFAULT '[]',
+    action_taken    VARCHAR(32)  NOT NULL,
+    create_time     TIMESTAMP    NOT NULL DEFAULT CURRENT_TIMESTAMP
+);
+
+CREATE INDEX IF NOT EXISTS idx_audit_log_created ON content_audit_log (create_time DESC);
+CREATE INDEX IF NOT EXISTS idx_audit_log_session ON content_audit_log (session_id);
+
+COMMENT ON TABLE  content_audit_log              IS '内容审计日志表（记录所有敏感词命中事件，只追加不删除）';
+COMMENT ON COLUMN content_audit_log.id           IS '主键（雪花算法生成）';
+COMMENT ON COLUMN content_audit_log.session_id   IS '会话ID（关联 conversation_session）';
+COMMENT ON COLUMN content_audit_log.direction    IS '检测方向: INPUT(用户输入) / OUTPUT(AI 输出)';
+COMMENT ON COLUMN content_audit_log.original_text IS '原始违规内容（截断至前 50 字）';
+COMMENT ON COLUMN content_audit_log.hit_words    IS '命中词列表（JSON 数组，记录触发的敏感词）';
+COMMENT ON COLUMN content_audit_log.action_taken IS '采取的动作: PASS(放行) / MASK(脱敏) / BLOCK(拦截)';
+COMMENT ON COLUMN content_audit_log.create_time  IS '创建时间（事件发生时间）';
+
+-- ============================================================
+-- 表 12: message_feedback — 消息反馈表（P0-002 用户反馈系统）
+-- ============================================================
+CREATE TABLE IF NOT EXISTS message_feedback (
+    id              BIGSERIAL    PRIMARY KEY,
+    message_id      VARCHAR(128) NOT NULL,
+    conversation_id VARCHAR(64)  NOT NULL,
+    feedback_type   VARCHAR(16)  NOT NULL,
+    reason_category VARCHAR(64),
+    reason_comment  TEXT,
+    create_time     TIMESTAMP    NOT NULL DEFAULT CURRENT_TIMESTAMP,
+    update_time     TIMESTAMP    NOT NULL DEFAULT CURRENT_TIMESTAMP,
+    is_delete       BOOLEAN      NOT NULL DEFAULT FALSE
+);
+
+-- 唯一索引：每条消息只能有一条反馈（重复提交覆盖，upsert 语义）
+CREATE UNIQUE INDEX IF NOT EXISTS uk_message_feedback_message ON message_feedback (message_id) WHERE is_delete = FALSE;
+CREATE INDEX IF NOT EXISTS idx_feedback_conversation ON message_feedback (conversation_id);
+CREATE INDEX IF NOT EXISTS idx_feedback_type         ON message_feedback (feedback_type) WHERE is_delete = FALSE;
+CREATE INDEX IF NOT EXISTS idx_feedback_created      ON message_feedback (create_time DESC);
+
+COMMENT ON TABLE  message_feedback                  IS '消息反馈表（用户对 AI 回复的点赞/点踩反馈）';
+COMMENT ON COLUMN message_feedback.id               IS '主键（雪花算法生成）';
+COMMENT ON COLUMN message_feedback.message_id       IS 'AI 消息ID（前端 msgId 或 UUID，唯一索引字段）';
+COMMENT ON COLUMN message_feedback.conversation_id  IS '会话ID（关联 chat_message）';
+COMMENT ON COLUMN message_feedback.feedback_type    IS '反馈类型: THUMBS_UP(有帮助) / THUMBS_DOWN(没帮助)';
+COMMENT ON COLUMN message_feedback.reason_category  IS '点踩原因分类: inaccurate / irrelevant / incomplete / other（仅 THUMBS_DOWN 时可选）';
+COMMENT ON COLUMN message_feedback.reason_comment   IS '自由文本补充说明（可选）';
+COMMENT ON COLUMN message_feedback.create_time      IS '创建时间';
+COMMENT ON COLUMN message_feedback.update_time      IS '更新时间（重复提交时覆盖）';
+COMMENT ON COLUMN message_feedback.is_delete        IS '逻辑删除: FALSE=正常 TRUE=已删除';
+
+-- ============================================================
+-- 表 13: knowledge_faq — FAQ 知识库表（P0-003 意图识别 + FAQ 精准匹配）
+-- ============================================================
+CREATE TABLE IF NOT EXISTS knowledge_faq (
+    id                  BIGSERIAL    PRIMARY KEY,
+    question            TEXT         NOT NULL,
+    answer              TEXT         NOT NULL,
+    similar_questions   TEXT         NOT NULL DEFAULT '[]',
+    category            VARCHAR(128),
+    status              VARCHAR(20)  NOT NULL DEFAULT 'ENABLED',
+    priority            INTEGER      NOT NULL DEFAULT 0,
+    hit_count           BIGINT       NOT NULL DEFAULT 0,
+    source              VARCHAR(64)  DEFAULT 'manual',
+    create_time         TIMESTAMP    NOT NULL DEFAULT CURRENT_TIMESTAMP,
+    update_time         TIMESTAMP    NOT NULL DEFAULT CURRENT_TIMESTAMP,
+    is_delete           BOOLEAN      NOT NULL DEFAULT FALSE
+);
+
+CREATE INDEX IF NOT EXISTS idx_faq_status   ON knowledge_faq (status) WHERE is_delete = FALSE;
+CREATE INDEX IF NOT EXISTS idx_faq_category ON knowledge_faq (category);
+CREATE INDEX IF NOT EXISTS idx_faq_priority ON knowledge_faq (priority DESC);
+
+COMMENT ON TABLE  knowledge_faq                      IS 'FAQ 知识库表（用于意图路由后的精准问答匹配）';
+COMMENT ON COLUMN knowledge_faq.id                   IS '主键（雪花算法生成）';
+COMMENT ON COLUMN knowledge_faq.question             IS '标准问题';
+COMMENT ON COLUMN knowledge_faq.answer               IS '标准答案';
+COMMENT ON COLUMN knowledge_faq.similar_questions    IS '相似问题列表（JSON 字符串数组，用于扩展匹配范围）';
+COMMENT ON COLUMN knowledge_faq.category             IS 'FAQ 分类';
+COMMENT ON COLUMN knowledge_faq.status               IS '状态: ENABLED(启用) / DISABLED(禁用)';
+COMMENT ON COLUMN knowledge_faq.priority             IS '优先级（数值越大越优先匹配，默认 0）';
+COMMENT ON COLUMN knowledge_faq.hit_count            IS '命中次数统计（用于分析高频问题）';
+COMMENT ON COLUMN knowledge_faq.source               IS '来源: manual(手动录入) / import(批量导入)';
+COMMENT ON COLUMN knowledge_faq.create_time          IS '创建时间';
+COMMENT ON COLUMN knowledge_faq.update_time          IS '更新时间';
+COMMENT ON COLUMN knowledge_faq.is_delete            IS '逻辑删除: FALSE=正常 TRUE=已删除';
+
+-- ============================================================
+-- 表 14: faq_embedding — FAQ 向量索引表（P0-003，FAQ 语义匹配用）
+-- ============================================================
+CREATE TABLE IF NOT EXISTS faq_embedding (
+    id              BIGSERIAL    PRIMARY KEY,
+    faq_id          BIGINT       NOT NULL,
+    embedding       vector(1024) NOT NULL,
+    model_name      VARCHAR(64)  NOT NULL,
+    create_time     TIMESTAMP    NOT NULL DEFAULT CURRENT_TIMESTAMP
+);
+
+CREATE INDEX IF NOT EXISTS idx_faq_emb_faq_id ON faq_embedding (faq_id);
+
+COMMENT ON TABLE  faq_embedding             IS 'FAQ 向量索引表（存储 FAQ 的语义向量，用于相似问题语义匹配）';
+COMMENT ON COLUMN faq_embedding.id          IS '主键';
+COMMENT ON COLUMN faq_embedding.faq_id      IS '关联 FAQ ID（对应 knowledge_faq.id）';
+COMMENT ON COLUMN faq_embedding.embedding   IS '向量嵌入（维度由 knowledge.vector.dimension 配置，默认 1024）';
+COMMENT ON COLUMN faq_embedding.model_name  IS '使用的 Embedding 模型名称（用于检测模型变更后重新计算）';
+COMMENT ON COLUMN faq_embedding.create_time IS '创建时间';
+
 -- ============================================================
 -- 完成！
--- 共创建 8 张表及对应索引。
+-- 共创建 14 张表及对应索引。
 -- ============================================================

src/main/resources/knowledge-base.sql

@@ -1,6 +1,7 @@
 -- ================================================================
 -- 知识库管理增强 - 数据库变更脚本
 -- 说明: 为知识库管理功能添加分类表和文档管理表
+-- 注意: 此脚本为早期版本，完整初始化请使用 init-database.sql
 -- ================================================================
 
 -- ================================================================
@@ -48,6 +49,7 @@ CREATE TABLE IF NOT EXISTS knowledge_document (
     chunk_count     INTEGER      DEFAULT 0             NOT NULL,
     status          VARCHAR(20)  DEFAULT 'PROCESSING'  NOT NULL,
     error_message   TEXT,
+    content_hash    VARCHAR(64),
     create_time     TIMESTAMP    DEFAULT CURRENT_TIMESTAMP NOT NULL,
     update_time     TIMESTAMP    DEFAULT CURRENT_TIMESTAMP NOT NULL,
     is_delete       BOOLEAN      DEFAULT FALSE         NOT NULL
@@ -68,6 +70,7 @@ COMMENT ON COLUMN knowledge_document.tags IS '标签列表(JSON数组)';
 COMMENT ON COLUMN knowledge_document.chunk_count IS '分块数量';
 COMMENT ON COLUMN knowledge_document.status IS '处理状态 - PROCESSING/READY/FAILED';
 COMMENT ON COLUMN knowledge_document.error_message IS '处理失败时的错误信息';
+COMMENT ON COLUMN knowledge_document.content_hash IS '内容 SHA-256 哈希值（用于文档去重校验）';
 COMMENT ON COLUMN knowledge_document.create_time IS '创建时间';
 COMMENT ON COLUMN knowledge_document.update_time IS '更新时间';
 COMMENT ON COLUMN knowledge_document.is_delete IS '删除标志 - false:未删除, true:已删除(逻辑删除)';
@@ -76,3 +79,4 @@ COMMENT ON COLUMN knowledge_document.is_delete IS '删除标志 - false:未删
 CREATE INDEX IF NOT EXISTS idx_knowledge_document_category ON knowledge_document (category_id);
 CREATE INDEX IF NOT EXISTS idx_knowledge_document_status ON knowledge_document (status);
 CREATE INDEX IF NOT EXISTS idx_knowledge_document_create_time ON knowledge_document (create_time DESC);
+CREATE INDEX IF NOT EXISTS idx_knowledge_document_content_hash ON knowledge_document (content_hash);

src/main/resources/static/components/ChatPanel.js

@@ -2,7 +2,7 @@
  * 智能客服对话面板
  */
 import { ref, computed, nextTick, onMounted } from 'vue'
-import { chatSync, chatRagSync, chatSSEUrl, chatRagSSEUrl, getRoleList, getAccountList, truncateConversation, ragSources } from '../js/api.js'
+import { chatSync, chatRagSync, chatSSEUrl, chatRagSSEUrl, getRoleList, getAccountList, truncateConversation, ragSources, submitFeedback as submitFeedbackApi } from '../js/api.js'
 import { toast, readSSEStream, renderMarkdown } from '../js/utils.js'
 import { store } from '../js/store.js'
 import MessageSources from './MessageSources.js'
@@ -124,6 +124,10 @@ export default {
                   <button v-if="m.role === 'assistant' && m.content && !m.streaming" @click="regenerate(i)">重新生成</button>
                   <button v-if="m.role === 'user' && !m.streaming" @click="startEditMessage(i)">编辑</button>
                   <button v-if="m.error" @click="retryLast">重试</button>
+                  <template v-if="m.role === 'assistant' && m.content && !m.streaming">
+                    <button @click="submitFeedback(m.id, 'up')" :style="{color: m.feedback === 'up' ? 'var(--primary)' : ''}" title="有帮助">👍</button>
+                    <button @click="submitFeedback(m.id, 'down')" :style="{color: m.feedback === 'down' ? '#dc3545' : ''}" title="没帮助">👎</button>
+                  </template>
                 </div>
               </template>
             </div>
@@ -201,6 +205,10 @@ export default {
       return new Date().toLocaleTimeString('zh-CN', { hour: '2-digit', minute: '2-digit' })
     }
 
+    function generateMsgId() {
+      return 'msg_' + Date.now() + '_' + Math.random().toString(36).substring(2, 8)
+    }
+
     function newChatId() {
       chatId.value = 'web_' + Date.now() + '_' + Math.random().toString(36).slice(2, 8)
     }
@@ -301,9 +309,9 @@ export default {
       userInput.value = ''
       lastUserInput.value = text
       isSending.value = true
-      messages.value.push({ role: 'user', content: text, streaming: false, time: formatTime() })
+      messages.value.push({ id: generateMsgId(), role: 'user', content: text, streaming: false, time: formatTime() })
 
-      const assistantMsg = { role: 'assistant', content: '', streaming: true, time: formatTime(), sources: [] }
+      const assistantMsg = { id: generateMsgId(), role: 'assistant', content: '', streaming: true, time: formatTime(), sources: [] }
       messages.value.push(assistantMsg)
       await scrollToBottom()
 
@@ -426,6 +434,30 @@ export default {
       }
     }
 
+    // ===== P0-002: 消息反馈 =====
+    async function submitFeedback(msgId, type) {
+      const msgIndex = messages.value.findIndex(m => m.id === msgId && m.role === 'assistant')
+      if (msgIndex === -1) return
+
+      const msg = messages.value[msgIndex]
+      const wasActive = msg.feedback === type
+      const newFeedback = wasActive ? null : type
+      msg.feedback = newFeedback
+
+      try {
+        await submitFeedbackApi({
+          messageId: String(msgId),
+          conversationId: chatId.value,
+          feedbackType: newFeedback ? (newFeedback === 'up' ? 'THUMBS_UP' : 'THUMBS_DOWN') : null
+        })
+        if (newFeedback) {
+          toast(newFeedback === 'up' ? '感谢反馈 👍' : '感谢反馈，我们会持续改进', 'success')
+        }
+      } catch (e) {
+        console.error('反馈提交失败:', e)
+      }
+    }
+
     onMounted(() => {
       newChatId()
       store.loadCategories()
@@ -464,7 +496,8 @@ export default {
       startEditMessage,
       cancelEdit,
       submitEdit,
-      regenerate
+      regenerate,
+      submitFeedback
     }
   }
 }

src/main/resources/static/components/DocSearch.js

@@ -1,5 +1,5 @@
 /**
- * 🔍 语义搜索测试
+ * 🔍 知识库搜索测试（支持多种检索模式）
  */
 import { ref } from 'vue'
 import { searchDocuments } from '../js/api.js'
@@ -8,15 +8,23 @@ import { toast } from '../js/utils.js'
 export default {
   template: `
     <div class="card">
-      <h2>语义搜索测试</h2>
-      <p style="color:var(--sub);font-size:13px;margin-bottom:12px;">输入查询语句，测试知识库检索效果</p>
+      <h2>知识库搜索测试</h2>
+      <p style="color:var(--sub);font-size:13px;margin-bottom:12px;">输入查询语句，测试知识库检索效果（支持向量、关键词、混合检索三种模式）</p>
       <div class="input-row">
         <input class="input" v-model="query" placeholder="输入查询，如：Spring Boot 是什么？" @keydown.enter="search">
+        <label class="input-field-sm">
+          <span>检索模式</span>
+          <select class="input input-sm" v-model="searchMode">
+            <option value="VECTOR">向量检索</option>
+            <option value="KEYWORD">关键词检索</option>
+            <option value="HYBRID">混合检索</option>
+          </select>
+        </label>
         <label class="input-field-sm">
           <span>返回条数</span>
           <input class="input input-sm" v-model.number="topK" type="number" min="1" max="20">
         </label>
-        <label class="input-field-sm">
+        <label v-if="searchMode !== 'KEYWORD'" class="input-field-sm">
           <span>相似度阈值</span>
           <input class="input input-sm" v-model.number="threshold" type="number" min="0" max="1" step="0.1">
         </label>
@@ -31,7 +39,11 @@ export default {
 
       <div v-else>
         <div v-for="(r, i) in results" :key="i" class="search-result">
-          <div class="score">相似度得分: {{ r.score !== null ? (1 - r.score).toFixed(4) : '-' }} | {{ r.title || '无标题' }} | {{ r.sourceName || '无来源' }}</div>
+          <div class="score">
+            <span style="display:inline-block;padding:1px 6px;border-radius:3px;font-size:11px;margin-right:6px;"
+                  :style="{background: modeColor(r.searchMode), color: '#fff'}">{{ modeLabel(r.searchMode) }}</span>
+            得分: {{ formatScore(r) }} | {{ r.title || '无标题' }} | {{ r.sourceName || '无来源' }}
+          </div>
           <div class="content">{{ r.content || '' }}</div>
         </div>
       </div>
@@ -41,6 +53,7 @@ export default {
     const query = ref('')
     const topK = ref(5)
     const threshold = ref(0.5)
+    const searchMode = ref('VECTOR')
     const results = ref([])
     const isSearching = ref(false)
     const searched = ref(false)
@@ -55,7 +68,7 @@ export default {
       searched.value = false
       errorMsg.value = ''
       try {
-        const json = await searchDocuments(query.value, topK.value, threshold.value)
+        const json = await searchDocuments(query.value, topK.value, threshold.value, null, null, searchMode.value)
         if (!json.success) {
           errorMsg.value = json.message
           results.value = []
@@ -72,6 +85,21 @@ export default {
       }
     }
 
-    return { query, topK, threshold, results, isSearching, searched, errorMsg, search }
+    function formatScore(r) {
+      if (r.searchMode === 'KEYWORD' || r.searchMode === 'HYBRID') {
+        return r.score != null ? r.score.toFixed(4) : '-'
+      }
+      return r.score != null ? (1 - r.score).toFixed(4) : '-'
+    }
+
+    function modeLabel(m) {
+      return { VECTOR: '向量', KEYWORD: '关键词', HYBRID: '混合' }[m] || m || '向量'
+    }
+
+    function modeColor(m) {
+      return { VECTOR: '#007bff', KEYWORD: '#28a745', HYBRID: '#6f42c1' }[m] || '#6c757d'
+    }
+
+    return { query, topK, threshold, searchMode, results, isSearching, searched, errorMsg, search, formatScore, modeLabel, modeColor }
   }
 }

src/main/resources/static/components/FaqManager.js

@@ -0,0 +1,299 @@
+/**
+ * FAQ 管理组件
+ * 支持 CRUD + 批量导入 + 导出 + 启用/禁用
+ */
+import { listFaqs, createFaq, updateFaq, deleteFaq, toggleFaqStatus, batchImportFaqs, exportFaqs, getFaqStats } from '../js/api.js'
+import { toast } from '../js/utils.js'
+
+export default {
+  template: `
+  <div class="card" style="margin-top:16px;">
+    <h3 style="margin-bottom:12px;">❓ FAQ 精准匹配管理</h3>
+
+    <!-- 操作栏 -->
+    <div style="display:flex;gap:8px;margin-bottom:12px;flex-wrap:wrap;align-items:center;">
+      <button @click="openAddDialog" style="padding:6px 14px;background:var(--primary);color:#fff;border:none;border-radius:6px;cursor:pointer;">＋ 添加 FAQ</button>
+      <button @click="showImportDialog = true" style="padding:6px 14px;background:#6c757d;color:#fff;border:none;border-radius:6px;cursor:pointer;">📥 批量导入</button>
+      <button @click="doExport" style="padding:6px 14px;background:#28a745;color:#fff;border:none;border-radius:6px;cursor:pointer;">📤 导出</button>
+      <input v-model="searchKeyword" @input="debouncedSearch" placeholder="搜索问题..." style="margin-left:auto;padding:6px 10px;border:1px solid var(--border);border-radius:6px;width:200px;" />
+      <select v-model="filterCategory" @change="loadList" style="padding:6px 10px;border:1px solid var(--border);border-radius:6px;">
+        <option value="">全部分类</option>
+        <option v-for="c in categories" :key="c" :value="c">{{ c }}</option>
+      </select>
+      <select v-model="filterStatus" @change="loadList" style="padding:6px 10px;border:1px solid var(--border);border-radius:6px;">
+        <option value="">全部状态</option>
+        <option value="ENABLED">启用</option>
+        <option value="DISABLED">禁用</option>
+      </select>
+    </div>
+
+    <!-- 统计概要 -->
+    <div v-if="stats" style="display:flex;gap:16px;margin-bottom:12px;font-size:13px;color:#666;">
+      <span>总计: {{ stats.totalCount || 0 }}</span>
+      <span>启用: {{ stats.enabledCount || 0 }}</span>
+      <span>总命中: {{ stats.totalHitCount || 0 }}</span>
+    </div>
+
+    <!-- FAQ 表格 -->
+    <table style="width:100%;border-collapse:collapse;">
+      <thead>
+        <tr style="border-bottom:2px solid var(--border);">
+          <th style="text-align:left;padding:8px;max-width:200px;">问题</th>
+          <th style="text-align:left;padding:8px;max-width:200px;">答案（摘要）</th>
+          <th style="text-align:center;padding:8px;">分类</th>
+          <th style="text-align:center;padding:8px;">优先级</th>
+          <th style="text-align:center;padding:8px;">命中</th>
+          <th style="text-align:center;padding:8px;">状态</th>
+          <th style="text-align:right;padding:8px;">操作</th>
+        </tr>
+      </thead>
+      <tbody>
+        <tr v-if="loading"><td colspan="7" style="text-align:center;padding:20px;">加载中...</td></tr>
+        <tr v-else-if="faqs.length === 0"><td colspan="7" style="text-align:center;padding:20px;color:#999;">暂无 FAQ</td></tr>
+        <tr v-for="f in faqs" :key="f.id" style="border-bottom:1px solid var(--border);">
+          <td style="padding:8px;max-width:200px;overflow:hidden;text-overflow:ellipsis;white-space:nowrap;" :title="f.question">{{ f.question }}</td>
+          <td style="padding:8px;max-width:200px;overflow:hidden;text-overflow:ellipsis;white-space:nowrap;color:#666;" :title="f.answer">{{ f.answer }}</td>
+          <td style="text-align:center;padding:8px;font-size:12px;">{{ f.category || '-' }}</td>
+          <td style="text-align:center;padding:8px;">{{ f.priority }}</td>
+          <td style="text-align:center;padding:8px;">{{ f.hitCount }}</td>
+          <td style="text-align:center;padding:8px;">
+            <span @click="toggleStatus(f)" style="cursor:pointer;" :style="{color: f.status === 'ENABLED' ? '#28a745' : '#dc3545'}">
+              {{ f.status === 'ENABLED' ? '✅ 启用' : '❌ 禁用' }}
+            </span>
+          </td>
+          <td style="text-align:right;padding:8px;">
+            <button @click="openEditDialog(f)" style="padding:3px 8px;background:none;border:1px solid var(--border);border-radius:4px;cursor:pointer;margin-right:4px;">编辑</button>
+            <button @click="removeFaq(f.id)" style="padding:3px 8px;background:none;border:1px solid #dc3545;color:#dc3545;border-radius:4px;cursor:pointer;">删除</button>
+          </td>
+        </tr>
+      </tbody>
+    </table>
+
+    <!-- 分页 -->
+    <div v-if="total > pageSize" style="display:flex;justify-content:center;gap:8px;margin-top:12px;">
+      <button @click="page > 1 && (page--, loadList())" :disabled="page <= 1" style="padding:4px 10px;">上一页</button>
+      <span style="line-height:32px;">第 {{ page }} / {{ Math.ceil(total / pageSize) }} 页（共 {{ total }} 条）</span>
+      <button @click="page < Math.ceil(total / pageSize) && (page++, loadList())" :disabled="page >= Math.ceil(total / pageSize)" style="padding:4px 10px;">下一页</button>
+    </div>
+
+    <!-- 新增/编辑弹窗 -->
+    <div v-if="showFormDialog" style="position:fixed;top:0;left:0;right:0;bottom:0;background:rgba(0,0,0,.5);display:flex;align-items:center;justify-content:center;z-index:1000;" @click.self="showFormDialog = false">
+      <div style="background:var(--card);border-radius:12px;padding:24px;width:560px;max-width:90vw;max-height:90vh;overflow-y:auto;">
+        <h4 style="margin-bottom:16px;">{{ editingFaq ? '编辑 FAQ' : '添加 FAQ' }}</h4>
+        <div style="margin-bottom:12px;">
+          <label style="display:block;margin-bottom:4px;font-size:13px;">问题 *</label>
+          <input v-model="form.question" placeholder="输入标准问题" style="width:100%;padding:8px;border:1px solid var(--border);border-radius:6px;box-sizing:border-box;" />
+        </div>
+        <div style="margin-bottom:12px;">
+          <label style="display:block;margin-bottom:4px;font-size:13px;">标准答案 *</label>
+          <textarea v-model="form.answer" rows="4" placeholder="输入标准答案" style="width:100%;padding:8px;border:1px solid var(--border);border-radius:6px;box-sizing:border-box;"></textarea>
+        </div>
+        <div style="margin-bottom:12px;">
+          <label style="display:block;margin-bottom:4px;font-size:13px;">相似问题（每行一个）</label>
+          <textarea v-model="form.similarQuestionsText" rows="3" placeholder="每行一个相似问法&#10;例如:&#10;怎么退货？&#10;退货流程是什么？" style="width:100%;padding:8px;border:1px solid var(--border);border-radius:6px;box-sizing:border-box;font-family:monospace;"></textarea>
+        </div>
+        <div style="display:flex;gap:8px;margin-bottom:12px;">
+          <div style="flex:1;">
+            <label style="display:block;margin-bottom:4px;font-size:13px;">分类</label>
+            <input v-model="form.category" placeholder="如: 退货政策" style="width:100%;padding:8px;border:1px solid var(--border);border-radius:6px;box-sizing:border-box;" />
+          </div>
+          <div style="width:100px;">
+            <label style="display:block;margin-bottom:4px;font-size:13px;">优先级</label>
+            <input v-model.number="form.priority" type="number" min="0" style="width:100%;padding:8px;border:1px solid var(--border);border-radius:6px;box-sizing:border-box;" />
+          </div>
+        </div>
+        <div style="display:flex;justify-content:flex-end;gap:8px;">
+          <button @click="showFormDialog = false" style="padding:8px 16px;background:none;border:1px solid var(--border);border-radius:6px;cursor:pointer;">取消</button>
+          <button @click="saveFaq" :disabled="!form.question || !form.answer" style="padding:8px 16px;background:var(--primary);color:#fff;border:none;border-radius:6px;cursor:pointer;">保存</button>
+        </div>
+      </div>
+    </div>
+
+    <!-- 批量导入弹窗 -->
+    <div v-if="showImportDialog" style="position:fixed;top:0;left:0;right:0;bottom:0;background:rgba(0,0,0,.5);display:flex;align-items:center;justify-content:center;z-index:1000;" @click.self="showImportDialog = false">
+      <div style="background:var(--card);border-radius:12px;padding:24px;width:560px;max-width:90vw;">
+        <h4 style="margin-bottom:16px;">批量导入 FAQ</h4>
+        <p style="font-size:13px;color:#666;margin-bottom:12px;">使用 JSON 格式批量导入，每条包含 question、answer、similarQuestions（可选数组）、category（可选）</p>
+        <textarea v-model="importJson" rows="10" placeholder='[&#10;  {"question":"退货流程","answer":"请先在订单页面...","similarQuestions":["怎么退货","退货步骤"],"category":"退货政策"},&#10;  {"question":"运费谁出","answer":"7天内退货运费..."}&#10;]' style="width:100%;padding:8px;border:1px solid var(--border);border-radius:6px;box-sizing:border-box;font-family:monospace;font-size:12px;"></textarea>
+        <div style="display:flex;justify-content:flex-end;gap:8px;margin-top:12px;">
+          <button @click="showImportDialog = false" style="padding:8px 16px;background:none;border:1px solid var(--border);border-radius:6px;cursor:pointer;">取消</button>
+          <button @click="doImport" :disabled="!importJson.trim()" style="padding:8px 16px;background:var(--primary);color:#fff;border:none;border-radius:6px;cursor:pointer;">导入</button>
+        </div>
+      </div>
+    </div>
+  </div>
+  `,
+
+  data() {
+    return {
+      faqs: [],
+      loading: false,
+      page: 1,
+      pageSize: 20,
+      total: 0,
+      searchKeyword: '',
+      filterCategory: '',
+      filterStatus: '',
+      categories: [],
+      stats: null,
+      showFormDialog: false,
+      showImportDialog: false,
+      editingFaq: null,
+      form: { question: '', answer: '', similarQuestionsText: '', category: '', priority: 0 },
+      importJson: '',
+      searchTimer: null
+    }
+  },
+
+  mounted() {
+    this.loadList()
+    this.loadStats()
+  },
+
+  methods: {
+    async loadList() {
+      this.loading = true
+      try {
+        const res = await listFaqs(this.page, this.pageSize, this.searchKeyword || undefined, this.filterCategory || undefined, this.filterStatus || undefined)
+        if (res.success) {
+          this.faqs = res.data?.records || res.data || []
+          this.total = res.data?.total || res.total || 0
+          // 提取分类列表
+          const cats = new Set(this.faqs.map(f => f.category).filter(Boolean))
+          this.categories = [...cats]
+        }
+      } catch (e) {
+        toast('加载失败: ' + e.message, 'error')
+      }
+      this.loading = false
+    },
+
+    async loadStats() {
+      try {
+        const res = await getFaqStats()
+        if (res.success) this.stats = res.data
+      } catch { /* silent */ }
+    },
+
+    debouncedSearch() {
+      clearTimeout(this.searchTimer)
+      this.searchTimer = setTimeout(() => { this.page = 1; this.loadList() }, 300)
+    },
+
+    openAddDialog() {
+      this.editingFaq = null
+      this.form = { question: '', answer: '', similarQuestionsText: '', category: '', priority: 0 }
+      this.showFormDialog = true
+    },
+
+    openEditDialog(f) {
+      this.editingFaq = f
+      let similarText = ''
+      try {
+        const arr = typeof f.similarQuestions === 'string' ? JSON.parse(f.similarQuestions) : f.similarQuestions
+        similarText = Array.isArray(arr) ? arr.join('\n') : ''
+      } catch { similarText = f.similarQuestions || '' }
+      this.form = { question: f.question, answer: f.answer, similarQuestionsText: similarText, category: f.category || '', priority: f.priority || 0 }
+      this.showFormDialog = true
+    },
+
+    async saveFaq() {
+      const similarQuestions = this.form.similarQuestionsText.split('\n').map(s => s.trim()).filter(Boolean)
+      const data = {
+        question: this.form.question,
+        answer: this.form.answer,
+        similarQuestions: JSON.stringify(similarQuestions),
+        category: this.form.category || null,
+        priority: this.form.priority || 0
+      }
+      try {
+        let res
+        if (this.editingFaq) {
+          res = await updateFaq(this.editingFaq.id, data)
+        } else {
+          res = await createFaq(data)
+        }
+        if (res.success) {
+          toast(this.editingFaq ? '修改成功' : '添加成功', 'success')
+          this.showFormDialog = false
+          this.loadList()
+          this.loadStats()
+        } else {
+          toast(res.message || '操作失败', 'error')
+        }
+      } catch (e) {
+        toast('操作失败: ' + e.message, 'error')
+      }
+    },
+
+    async toggleStatus(f) {
+      const newStatus = f.status === 'ENABLED' ? 'DISABLED' : 'ENABLED'
+      try {
+        const res = await toggleFaqStatus(f.id, newStatus)
+        if (res.success) {
+          f.status = newStatus
+          toast(`已${newStatus === 'ENABLED' ? '启用' : '禁用'}`, 'success')
+        } else {
+          toast(res.message || '操作失败', 'error')
+        }
+      } catch (e) {
+        toast('操作失败: ' + e.message, 'error')
+      }
+    },
+
+    async removeFaq(id) {
+      if (!confirm('确认删除该 FAQ？')) return
+      try {
+        const res = await deleteFaq(id)
+        if (res.success) { toast('删除成功', 'success'); this.loadList(); this.loadStats() }
+        else toast(res.message || '删除失败', 'error')
+      } catch (e) {
+        toast('删除失败: ' + e.message, 'error')
+      }
+    },
+
+    async doImport() {
+      try {
+        const faqs = JSON.parse(this.importJson)
+        if (!Array.isArray(faqs) || faqs.length === 0) return toast('请输入有效的 FAQ 数组', 'error')
+        // 处理 similarQuestions 字段
+        for (const f of faqs) {
+          if (Array.isArray(f.similarQuestions)) f.similarQuestions = JSON.stringify(f.similarQuestions)
+          else if (!f.similarQuestions) f.similarQuestions = '[]'
+        }
+        const res = await batchImportFaqs({ faqs })
+        if (res.success) {
+          toast(`成功导入 ${res.data || faqs.length} 条`, 'success')
+          this.showImportDialog = false
+          this.importJson = ''
+          this.loadList()
+          this.loadStats()
+        } else {
+          toast(res.message || '导入失败', 'error')
+        }
+      } catch (e) {
+        if (e instanceof SyntaxError) toast('JSON 格式错误', 'error')
+        else toast('导入失败: ' + e.message, 'error')
+      }
+    },
+
+    async doExport() {
+      try {
+        const res = await exportFaqs()
+        if (res.success) {
+          const data = JSON.stringify(res.data, null, 2)
+          const blob = new Blob([data], { type: 'application/json' })
+          const url = URL.createObjectURL(blob)
+          const a = document.createElement('a')
+          a.href = url; a.download = 'faq_export.json'; a.click()
+          URL.revokeObjectURL(url)
+          toast('导出成功', 'success')
+        } else {
+          toast(res.message || '导出失败', 'error')
+        }
+      } catch (e) {
+        toast('导出失败: ' + e.message, 'error')
+      }
+    }
+  }
+}

src/main/resources/static/components/SensitiveWordManager.js

@@ -0,0 +1,304 @@
+/**
+ * 敏感词管理组件
+ * 支持 CRUD + 批量导入 + 审计日志查看
+ */
+import { listSensitiveWords, createSensitiveWord, updateSensitiveWord, deleteSensitiveWord, batchImportSensitiveWords, listAuditLogs } from '../js/api.js'
+import { toast } from '../js/utils.js'
+
+export default {
+  template: `
+  <div class="card" style="margin-top:16px;">
+    <h3 style="margin-bottom:12px;">🛡️ 敏感词管理</h3>
+
+    <!-- 操作栏 -->
+    <div style="display:flex;gap:8px;margin-bottom:12px;flex-wrap:wrap;align-items:center;">
+      <button @click="showAddDialog = true" style="padding:6px 14px;background:var(--primary);color:#fff;border:none;border-radius:6px;cursor:pointer;">＋ 添加敏感词</button>
+      <button @click="showImportDialog = true" style="padding:6px 14px;background:#6c757d;color:#fff;border:none;border-radius:6px;cursor:pointer;">📥 批量导入</button>
+      <button @click="showAuditLog = !showAuditLog" style="padding:6px 14px;background:#17a2b8;color:#fff;border:none;border-radius:6px;cursor:pointer;">📋 审计日志</button>
+      <input v-model="searchKeyword" @input="debouncedSearch" placeholder="搜索敏感词..." style="margin-left:auto;padding:6px 10px;border:1px solid var(--border);border-radius:6px;width:200px;" />
+      <select v-model="filterCategory" @change="loadList" style="padding:6px 10px;border:1px solid var(--border);border-radius:6px;">
+        <option value="">全部分类</option>
+        <option value="politics">政治</option>
+        <option value="porn">色情</option>
+        <option value="abuse">辱骂</option>
+        <option value="custom">自定义</option>
+      </select>
+    </div>
+
+    <!-- 敏感词表格 -->
+    <table style="width:100%;border-collapse:collapse;">
+      <thead>
+        <tr style="border-bottom:2px solid var(--border);">
+          <th style="text-align:left;padding:8px;">敏感词</th>
+          <th style="text-align:left;padding:8px;">分类</th>
+          <th style="text-align:center;padding:8px;">级别</th>
+          <th style="text-align:center;padding:8px;">状态</th>
+          <th style="text-align:right;padding:8px;">操作</th>
+        </tr>
+      </thead>
+      <tbody>
+        <tr v-if="loading"><td colspan="5" style="text-align:center;padding:20px;">加载中...</td></tr>
+        <tr v-else-if="words.length === 0"><td colspan="5" style="text-align:center;padding:20px;color:#999;">暂无数据</td></tr>
+        <tr v-for="w in words" :key="w.id" style="border-bottom:1px solid var(--border);">
+          <td style="padding:8px;">{{ w.word }}</td>
+          <td style="padding:8px;"><span style="padding:2px 8px;border-radius:4px;font-size:12px;" :style="{background: categoryColor(w.category)}">{{ categoryLabel(w.category) }}</span></td>
+          <td style="text-align:center;padding:8px;"><span :style="{color: w.level >= 2 ? '#dc3545' : '#ffc107'}">{{ w.level >= 2 ? '拦截' : '警告' }}</span></td>
+          <td style="text-align:center;padding:8px;">{{ w.isActive ? '✅' : '❌' }}</td>
+          <td style="text-align:right;padding:8px;">
+            <button @click="editWord(w)" style="padding:3px 8px;background:none;border:1px solid var(--border);border-radius:4px;cursor:pointer;margin-right:4px;">编辑</button>
+            <button @click="removeWord(w.id)" style="padding:3px 8px;background:none;border:1px solid #dc3545;color:#dc3545;border-radius:4px;cursor:pointer;">删除</button>
+          </td>
+        </tr>
+      </tbody>
+    </table>
+
+    <!-- 分页 -->
+    <div v-if="total > pageSize" style="display:flex;justify-content:center;gap:8px;margin-top:12px;">
+      <button @click="page > 1 && (page--, loadList())" :disabled="page <= 1" style="padding:4px 10px;">上一页</button>
+      <span style="line-height:32px;">第 {{ page }} / {{ Math.ceil(total / pageSize) }} 页（共 {{ total }} 条）</span>
+      <button @click="page < Math.ceil(total / pageSize) && (page++, loadList())" :disabled="page >= Math.ceil(total / pageSize)" style="padding:4px 10px;">下一页</button>
+    </div>
+
+    <!-- 新增/编辑弹窗 -->
+    <div v-if="showAddDialog || editingWord" style="position:fixed;top:0;left:0;right:0;bottom:0;background:rgba(0,0,0,.5);display:flex;align-items:center;justify-content:center;z-index:1000;" @click.self="closeDialog">
+      <div style="background:var(--card);border-radius:12px;padding:24px;width:420px;max-width:90vw;">
+        <h4 style="margin-bottom:16px;">{{ editingWord ? '编辑敏感词' : '添加敏感词' }}</h4>
+        <div style="margin-bottom:12px;">
+          <label style="display:block;margin-bottom:4px;font-size:13px;">敏感词</label>
+          <input v-model="form.word" placeholder="输入敏感词" style="width:100%;padding:8px;border:1px solid var(--border);border-radius:6px;box-sizing:border-box;" />
+        </div>
+        <div style="margin-bottom:12px;">
+          <label style="display:block;margin-bottom:4px;font-size:13px;">分类</label>
+          <select v-model="form.category" style="width:100%;padding:8px;border:1px solid var(--border);border-radius:6px;">
+            <option value="custom">自定义</option>
+            <option value="politics">政治</option>
+            <option value="porn">色情</option>
+            <option value="abuse">辱骂</option>
+          </select>
+        </div>
+        <div style="margin-bottom:12px;">
+          <label style="display:block;margin-bottom:4px;font-size:13px;">级别</label>
+          <select v-model="form.level" style="width:100%;padding:8px;border:1px solid var(--border);border-radius:6px;">
+            <option :value="1">警告（脱敏处理）</option>
+            <option :value="2">拦截（阻断回复）</option>
+          </select>
+        </div>
+        <div style="margin-bottom:16px;">
+          <label style="display:block;margin-bottom:4px;font-size:13px;">备注</label>
+          <input v-model="form.remark" placeholder="可选备注" style="width:100%;padding:8px;border:1px solid var(--border);border-radius:6px;box-sizing:border-box;" />
+        </div>
+        <div style="display:flex;justify-content:flex-end;gap:8px;">
+          <button @click="closeDialog" style="padding:8px 16px;background:none;border:1px solid var(--border);border-radius:6px;cursor:pointer;">取消</button>
+          <button @click="saveWord" :disabled="!form.word" style="padding:8px 16px;background:var(--primary);color:#fff;border:none;border-radius:6px;cursor:pointer;">保存</button>
+        </div>
+      </div>
+    </div>
+
+    <!-- 批量导入弹窗 -->
+    <div v-if="showImportDialog" style="position:fixed;top:0;left:0;right:0;bottom:0;background:rgba(0,0,0,.5);display:flex;align-items:center;justify-content:center;z-index:1000;" @click.self="showImportDialog = false">
+      <div style="background:var(--card);border-radius:12px;padding:24px;width:500px;max-width:90vw;">
+        <h4 style="margin-bottom:16px;">批量导入敏感词</h4>
+        <div style="margin-bottom:12px;">
+          <label style="display:block;margin-bottom:4px;font-size:13px;">每行一个敏感词</label>
+          <textarea v-model="importText" rows="8" placeholder="每行输入一个敏感词&#10;例如:&#10;违禁词1&#10;违禁词2" style="width:100%;padding:8px;border:1px solid var(--border);border-radius:6px;box-sizing:border-box;font-family:monospace;"></textarea>
+        </div>
+        <div style="display:flex;gap:8px;margin-bottom:16px;">
+          <select v-model="importCategory" style="padding:8px;border:1px solid var(--border);border-radius:6px;">
+            <option value="custom">自定义</option>
+            <option value="politics">政治</option>
+            <option value="porn">色情</option>
+            <option value="abuse">辱骂</option>
+          </select>
+          <select v-model="importLevel" style="padding:8px;border:1px solid var(--border);border-radius:6px;">
+            <option :value="1">警告</option>
+            <option :value="2">拦截</option>
+          </select>
+        </div>
+        <div style="display:flex;justify-content:flex-end;gap:8px;">
+          <button @click="showImportDialog = false" style="padding:8px 16px;background:none;border:1px solid var(--border);border-radius:6px;cursor:pointer;">取消</button>
+          <button @click="doImport" :disabled="!importText.trim()" style="padding:8px 16px;background:var(--primary);color:#fff;border:none;border-radius:6px;cursor:pointer;">导入</button>
+        </div>
+      </div>
+    </div>
+
+    <!-- 审计日志面板 -->
+    <div v-if="showAuditLog" ref="auditLogPanel" style="margin-top:16px;border-top:1px solid var(--border);padding-top:16px;">
+      <h4 style="margin-bottom:12px;">📋 内容审计日志</h4>
+      <table style="width:100%;border-collapse:collapse;font-size:13px;">
+        <thead>
+          <tr style="border-bottom:2px solid var(--border);">
+            <th style="text-align:left;padding:6px;">时间</th>
+            <th style="text-align:center;padding:6px;">方向</th>
+            <th style="text-align:left;padding:6px;">违规内容</th>
+            <th style="text-align:left;padding:6px;">命中词</th>
+            <th style="text-align:center;padding:6px;">处理</th>
+          </tr>
+        </thead>
+        <tbody>
+          <tr v-if="auditLogs.length === 0"><td colspan="5" style="text-align:center;padding:12px;color:#999;">暂无审计日志</td></tr>
+          <tr v-for="log in auditLogs" :key="log.id" style="border-bottom:1px solid var(--border);">
+            <td style="padding:6px;white-space:nowrap;">{{ formatTime(log.createTime) }}</td>
+            <td style="text-align:center;padding:6px;"><span :style="{color: log.direction === 'INPUT' ? '#007bff' : '#28a745'}">{{ log.direction === 'INPUT' ? '输入' : '输出' }}</span></td>
+            <td style="padding:6px;max-width:200px;overflow:hidden;text-overflow:ellipsis;white-space:nowrap;" :title="log.originalText">{{ log.originalText }}</td>
+            <td style="padding:6px;font-size:12px;">{{ formatHitWords(log.hitWords) }}</td>
+            <td style="text-align:center;padding:6px;">
+              <span :style="{color: log.actionTaken === 'BLOCK' ? '#dc3545' : log.actionTaken === 'MASK' ? '#ffc107' : '#28a745'}">
+                {{ log.actionTaken === 'BLOCK' ? '拦截' : log.actionTaken === 'MASK' ? '脱敏' : '通过' }}
+              </span>
+            </td>
+          </tr>
+        </tbody>
+      </table>
+    </div>
+  </div>
+  `,
+
+  data() {
+    return {
+      words: [],
+      loading: false,
+      page: 1,
+      pageSize: 20,
+      total: 0,
+      searchKeyword: '',
+      filterCategory: '',
+      showAddDialog: false,
+      showImportDialog: false,
+      showAuditLog: false,
+      editingWord: null,
+      form: { word: '', category: 'custom', level: 1, remark: '' },
+      importText: '',
+      importCategory: 'custom',
+      importLevel: 1,
+      auditLogs: [],
+      searchTimer: null
+    }
+  },
+
+  mounted() {
+    this.loadList()
+  },
+
+  methods: {
+    async loadList() {
+      this.loading = true
+      try {
+        const res = await listSensitiveWords(this.page, this.pageSize, this.searchKeyword || undefined, this.filterCategory || undefined)
+        if (res.success) {
+          this.words = res.data?.records || res.data || []
+          this.total = res.data?.total || res.total || 0
+        }
+      } catch (e) {
+        toast('加载失败: ' + e.message, 'error')
+      }
+      this.loading = false
+    },
+
+    debouncedSearch() {
+      clearTimeout(this.searchTimer)
+      this.searchTimer = setTimeout(() => { this.page = 1; this.loadList() }, 300)
+    },
+
+    editWord(w) {
+      this.editingWord = w
+      this.form = { word: w.word, category: w.category, level: w.level, remark: w.remark || '' }
+    },
+
+    closeDialog() {
+      this.showAddDialog = false
+      this.editingWord = null
+      this.form = { word: '', category: 'custom', level: 1, remark: '' }
+    },
+
+    async saveWord() {
+      try {
+        if (this.editingWord) {
+          const res = await updateSensitiveWord(this.editingWord.id, this.form)
+          if (res.success) toast('修改成功', 'success')
+          else toast(res.message || '修改失败', 'error')
+        } else {
+          const res = await createSensitiveWord(this.form)
+          if (res.success) toast('添加成功', 'success')
+          else toast(res.message || '添加失败', 'error')
+        }
+        this.closeDialog()
+        this.loadList()
+      } catch (e) {
+        toast('操作失败: ' + e.message, 'error')
+      }
+    },
+
+    async removeWord(id) {
+      if (!confirm('确认删除该敏感词？')) return
+      try {
+        const res = await deleteSensitiveWord(id)
+        if (res.success) { toast('删除成功', 'success'); this.loadList() }
+        else toast(res.message || '删除失败', 'error')
+      } catch (e) {
+        toast('删除失败: ' + e.message, 'error')
+      }
+    },
+
+    async doImport() {
+      const words = this.importText.split('\n').map(s => s.trim()).filter(Boolean)
+      if (words.length === 0) return toast('请输入敏感词', 'error')
+      try {
+        const res = await batchImportSensitiveWords({ words, category: this.importCategory, level: this.importLevel })
+        if (res.success) {
+          toast(`成功导入 ${res.data || words.length} 条`, 'success')
+          this.showImportDialog = false
+          this.importText = ''
+          this.loadList()
+        } else {
+          toast(res.message || '导入失败', 'error')
+        }
+      } catch (e) {
+        toast('导入失败: ' + e.message, 'error')
+      }
+    },
+
+    async loadAuditLogs() {
+      try {
+        const res = await listAuditLogs(1, 50)
+        if (res.success) {
+          this.auditLogs = res.data?.records || res.data || []
+        } else {
+          toast('加载审计日志失败: ' + (res.message || '未知错误'), 'error')
+        }
+      } catch (e) {
+        toast('加载审计日志失败: ' + e.message, 'error')
+      }
+    },
+
+    categoryLabel(c) { return { politics: '政治', porn: '色情', abuse: '辱骂', custom: '自定义' }[c] || c },
+    categoryColor(c) { return { politics: '#dc3545', porn: '#e83e8c', abuse: '#fd7e14', custom: '#6c757d' }[c] || '#6c757d' },
+    formatTime(t) { return t ? new Date(t).toLocaleString('zh-CN') : '' },
+    formatHitWords(hw) {
+      if (!hw) return ''
+      // PostgreSQL JDBC 返回 JSONB 为 PGobject：{"type":"jsonb","value":"...", "null":false}
+      if (hw.type === 'jsonb' && hw.value) {
+        try { hw = JSON.parse(hw.value) } catch { return hw.value }
+      }
+      if (typeof hw === 'string') { try { hw = JSON.parse(hw) } catch { return hw } }
+      // 格式：{"hits":[{"word":"xxx","level":2,"category":"politics"}]}
+      if (hw && hw.hits && Array.isArray(hw.hits)) {
+        return hw.hits.map(h => `${h.word || ''}(${h.category || ''})`).join('、')
+      }
+      if (Array.isArray(hw)) return hw.map(h => typeof h === 'string' ? h : h.word || '').join('、')
+      return String(hw)
+    }
+  },
+
+  watch: {
+    showAuditLog(val) {
+      if (val) {
+        this.loadAuditLogs()
+        this.$nextTick(() => {
+          if (this.$refs.auditLogPanel) {
+            this.$refs.auditLogPanel.scrollIntoView({ behavior: 'smooth', block: 'start' })
+          }
+        })
+      }
+    }
+  }
+}

src/main/resources/static/js/api.js

@@ -274,10 +274,11 @@ export function batchReprocessDocuments(ids) {
 /**
  * 语义搜索
  */
-export function searchDocuments(query, topK, similarityThreshold, categoryId, categoryIds) {
+export function searchDocuments(query, topK, similarityThreshold, categoryId, categoryIds, searchMode) {
   const body = { query, topK, similarityThreshold }
   if (categoryIds && categoryIds.length) body.categoryIds = categoryIds
   if (categoryId) body.categoryId = categoryId
+  if (searchMode) body.searchMode = searchMode
   return postJSON('/document/search', body)
 }
 
@@ -507,3 +508,150 @@ export function deleteModelConfig(id) {
 export function truncateConversation(conversationId, userTurn) {
   return postJSON(`/conversation/${conversationId}/truncate`, { userTurn })
 }
+
+// ==================== P0-002: 用户反馈 ====================
+
+/**
+ * 提交/修改消息反馈（upsert 语义）
+ * @param {Object} feedback - { messageId, conversationId, feedbackType, reasonCategory, reasonComment }
+ */
+export function submitFeedback(feedback) {
+  return postJSON('/feedback', feedback)
+}
+
+/**
+ * 获取反馈统计
+ */
+export function getFeedbackStats(startDate, endDate) {
+  let path = '/feedback/stats'
+  const params = []
+  if (startDate) params.push(`startDate=${encodeURIComponent(startDate)}`)
+  if (endDate) params.push(`endDate=${encodeURIComponent(endDate)}`)
+  if (params.length) path += '?' + params.join('&')
+  return getJSON(path)
+}
+
+/**
+ * 按会话查询反馈
+ */
+export function getFeedbackByConversation(conversationId) {
+  return getJSON(`/feedback/by-conversation/${encodeURIComponent(conversationId)}`)
+}
+
+/**
+ * 批量查询消息反馈状态（供回显）
+ */
+export function getFeedbackBatch(messageIds) {
+  return getJSON(`/feedback/batch?messageIds=${encodeURIComponent(messageIds.join(','))}`)
+}
+
+// ==================== P0-004: 敏感词管理 ====================
+
+/**
+ * 敏感词分页列表
+ */
+export function listSensitiveWords(page = 1, size = 20, keyword, category) {
+  let path = `/sensitive-word/list?page=${page}&size=${size}`
+  if (keyword) path += `&keyword=${encodeURIComponent(keyword)}`
+  if (category) path += `&category=${encodeURIComponent(category)}`
+  return getJSON(path)
+}
+
+/**
+ * 新增敏感词
+ */
+export function createSensitiveWord(data) {
+  return postJSON('/sensitive-word', data)
+}
+
+/**
+ * 修改敏感词
+ */
+export function updateSensitiveWord(id, data) {
+  return putJSONWithBody(`/sensitive-word/${id}`, data)
+}
+
+/**
+ * 删除敏感词
+ */
+export function deleteSensitiveWord(id) {
+  return deleteJSON(`/sensitive-word/${id}`)
+}
+
+/**
+ * 批量导入敏感词
+ */
+export function batchImportSensitiveWords(data) {
+  return postJSON('/sensitive-word/batch-import', data)
+}
+
+/**
+ * 查询审计日志
+ */
+export function listAuditLogs(page = 1, size = 20, sessionId) {
+  let path = `/sensitive-word/audit-log?page=${page}&size=${size}`
+  if (sessionId) path += `&sessionId=${encodeURIComponent(sessionId)}`
+  return getJSON(path)
+}
+
+// ==================== P0-003: FAQ 管理 ====================
+
+/**
+ * FAQ 分页列表
+ */
+export function listFaqs(page = 1, size = 20, keyword, category, status) {
+  let path = `/faq/list?page=${page}&size=${size}`
+  if (keyword) path += `&keyword=${encodeURIComponent(keyword)}`
+  if (category) path += `&category=${encodeURIComponent(category)}`
+  if (status) path += `&status=${encodeURIComponent(status)}`
+  return getJSON(path)
+}
+
+/**
+ * 新增 FAQ
+ */
+export function createFaq(data) {
+  return postJSON('/faq', data)
+}
+
+/**
+ * 修改 FAQ
+ */
+export function updateFaq(id, data) {
+  return putJSONWithBody(`/faq/${id}`, data)
+}
+
+/**
+ * 删除 FAQ
+ */
+export function deleteFaq(id) {
+  return deleteJSON(`/faq/${id}`)
+}
+
+/**
+ * 启用/禁用 FAQ
+ */
+export function toggleFaqStatus(id, status) {
+  return putJSON(`/faq/${id}/toggle`, { status })
+}
+
+/**
+ * 批量导入 FAQ
+ */
+export function batchImportFaqs(data) {
+  return postJSON('/faq/batch-import', data)
+}
+
+/**
+ * 导出所有 FAQ
+ */
+export function exportFaqs() {
+  return getJSON('/faq/export')
+}
+
+/**
+ * FAQ 匹配统计
+ */
+export function getFaqStats() {
+  return getJSON('/faq/stats')
+}

src/main/resources/static/js/app.js

@@ -17,6 +17,8 @@ import DocUpload from '../components/DocUpload.js'
 import DocDetail from '../components/DocDetail.js'
 import ConversationManager from '../components/ConversationManager.js'
 import ModelConfigManager from '../components/ModelConfigManager.js'
+import SensitiveWordManager from '../components/SensitiveWordManager.js'
+import FaqManager from '../components/FaqManager.js'
 
 const app = createApp({
   setup() {
@@ -75,6 +77,7 @@ const app = createApp({
         <category-manager></category-manager>
         <doc-list></doc-list>
         <doc-upload></doc-upload>
+        <faq-manager></faq-manager>
       </div>
 
       <!-- Tab 3: 会话管理 -->
@@ -87,6 +90,7 @@ const app = createApp({
         <account-manager></account-manager>
         <role-manager></role-manager>
         <model-config-manager></model-config-manager>
+        <sensitive-word-manager></sensitive-word-manager>
       </div>
 
       <!-- 文档详情弹窗 -->
@@ -110,5 +114,7 @@ app.component('doc-upload', DocUpload)
 app.component('doc-detail', DocDetail)
 app.component('conversation-manager', ConversationManager)
 app.component('model-config-manager', ModelConfigManager)
+app.component('sensitive-word-manager', SensitiveWordManager)
+app.component('faq-manager', FaqManager)
 
 app.mount('#app')

src/main/resources/static/sdk/chatbot-sdk.js

@@ -755,6 +755,36 @@ var ChatbotSDK = (function () {
             return [];
         }
     }
+    // ==================== P0-002: 消息反馈 ====================
+    /**
+     * 提交消息反馈（点赞/点踩）
+     */
+    async function submitFeedbackApi(messageId, feedbackType) {
+        if (!currentConfig)
+            return false;
+        const url = buildUrl('/feedback');
+        try {
+            const response = await safeFetch(url, {
+                method: 'POST',
+                headers: { 'Content-Type': 'application/json' },
+                body: JSON.stringify({
+                    messageId: String(messageId),
+                    conversationId: currentConfig.chatId,
+                    feedbackType,
+                }),
+            });
+            if (!response.ok) {
+                logger.error(`反馈提交失败 status=${response.status}`);
+                return false;
+            }
+            const json = await response.json();
+            return json.success || false;
+        }
+        catch (err) {
+            logger.error('反馈提交异常', err);
+            return false;
+        }
+    }
     /**
      * 获取会话列表
      */
@@ -3654,8 +3684,7 @@ var ChatbotSDK = (function () {
     // ==================== 消息反馈（👍 / 👎） ====================
     /**
      * 处理消息反馈：切换 AI 消息的点赞/点踩状态
-     * 当前仅前端状态持久化（存入 messages 数组 + localStorage），
-     * 预留后端对接位：TODO 接口 POST /conversation/message/feedback
+     * 前端状态持久化（存入 messages 数组 + localStorage）+ 调用后端 API 记录反馈
      */
     function handleFeedback(msgId, value) {
         if (!messagesContainer$1)
@@ -3673,7 +3702,21 @@ var ChatbotSDK = (function () {
         // 持久化（localStorage）
         if (config$1)
             saveMessages(config$1.integrateId, messages);
-        logger.info(`消息反馈 msgId=${msgId} value=${newValue || 'cleared'}`);
+        // 调用后端 API 记录反馈
+        if (newValue) {
+            const feedbackType = newValue === 'up' ? 'THUMBS_UP' : 'THUMBS_DOWN';
+            submitFeedbackApi(String(msgId), feedbackType).then(success => {
+                if (success) {
+                    logger.info(`消息反馈已提交 msgId=${msgId} value=${newValue}`);
+                }
+                else {
+                    logger.warn(`消息反馈提交失败 msgId=${msgId}（本地状态已更新）`);
+                }
+            });
+        }
+        else {
+            logger.info(`消息反馈已取消 msgId=${msgId}`);
+        }
     }
     function autoResizeInput() {
         if (!inputEl$1)