添加重排序功能

ruoyi-modules/ruoyi-chat/src/main/java/org/ruoyi/domain/bo/knowledge/KnowledgeInfoBo.java

@@ -77,10 +77,33 @@ public class KnowledgeInfoBo extends BaseEntity {
      */
     private String embeddingModel;
 
+    /**
+     * 是否启用重排序（0 否 1是）
+     */
+    private Integer enableRerank;
+
+    /**
+     * 重排序模型名称
+     */
+    private String rerankModel;
+
+    /**
+     * 重排序后返回的文档数量
+     */
+    private Integer rerankTopN;
+
+    /**
+     * 重排序相关性分数阈值
+     */
+    private Double rerankScoreThreshold;
+
+
     /**
      * 备注
      */
     private String remark;
 
 
+
+
 }

ruoyi-modules/ruoyi-chat/src/main/java/org/ruoyi/domain/bo/rerank/RerankRequest.java

@@ -0,0 +1,44 @@
+package org.ruoyi.domain.bo.rerank;
+
+import lombok.AllArgsConstructor;
+import lombok.Builder;
+import lombok.Data;
+import lombok.NoArgsConstructor;
+
+import java.util.List;
+
+/**
+ * 重排序请求参数
+ *
+ * @author yang
+ * @date 2026-04-19
+ */
+@Data
+@Builder
+@NoArgsConstructor
+@AllArgsConstructor
+public class RerankRequest {
+
+    /**
+     * 查询文本
+     */
+    private String query;
+
+    /**
+     * 候选文档列表
+     */
+    private List<String> documents;
+
+    /**
+     * 返回的文档数量（topN）
+     * 如果不指定，默认返回所有文档
+     */
+    private Integer topN;
+
+    /**
+     * 是否返回原始文档内容
+     * 默认为 true
+     */
+    @Builder.Default
+    private Boolean returnDocuments = true;
+}

ruoyi-modules/ruoyi-chat/src/main/java/org/ruoyi/domain/bo/rerank/RerankResult.java

@@ -0,0 +1,72 @@
+package org.ruoyi.domain.bo.rerank;
+
+import lombok.AllArgsConstructor;
+import lombok.Builder;
+import lombok.Data;
+import lombok.NoArgsConstructor;
+
+import java.util.List;
+
+/**
+ * 重排序结果
+ *
+ * @author yang
+ * @date 2026-04-19
+ */
+@Data
+@Builder
+@NoArgsConstructor
+@AllArgsConstructor
+public class RerankResult {
+
+    /**
+     * 重排序后的文档结果列表
+     */
+    private List<RerankDocument> documents;
+
+    /**
+     * 原始请求中的文档总数
+     */
+    private Integer totalDocuments;
+
+    /**
+     * 重排序耗时（毫秒）
+     */
+    private Long durationMs;
+
+    /**
+     * 单个重排序文档结果
+     */
+    @Data
+    @Builder
+    @NoArgsConstructor
+    @AllArgsConstructor
+    public static class RerankDocument {
+
+        /**
+         * 文档在原始列表中的索引位置
+         */
+        private Integer index;
+
+        /**
+         * 相关性分数（通常 0-1 之间，越高越相关）
+         */
+        private Double relevanceScore;
+
+        /**
+         * 文档内容
+         */
+        private String document;
+    }
+
+    /**
+     * 创建空结果
+     */
+    public static RerankResult empty() {
+        return RerankResult.builder()
+                .documents(List.of())
+                .totalDocuments(0)
+                .durationMs(0L)
+                .build();
+    }
+}

ruoyi-modules/ruoyi-chat/src/main/java/org/ruoyi/domain/bo/vector/QueryVectorBo.java

@@ -51,4 +51,30 @@ public class QueryVectorBo {
      */
     private String baseUrl;
 
+
+    // ========== 重排序相关参数 ==========
+
+    /**
+     * 是否启用重排序
+     * 默认为 false
+     */
+    private Boolean enableRerank = false;
+
+    /**
+     * 重排序模型名称
+     */
+    private String rerankModelName;
+
+    /**
+     * 重排序后返回的文档数量（topN）
+     * 如果不指定，默认与 maxResults 相同
+     */
+    private Integer rerankTopN;
+
+    /**
+     * 重排序相关性分数阈值
+     * 低于此阈值的文档将被过滤
+     */
+    private Double rerankScoreThreshold;
+
 }

ruoyi-modules/ruoyi-chat/src/main/java/org/ruoyi/domain/dto/request/ZhipuRerankRequest.java

@@ -0,0 +1,48 @@
+package org.ruoyi.domain.dto.request;
+
+import com.fasterxml.jackson.core.JsonProcessingException;
+import com.fasterxml.jackson.databind.ObjectMapper;
+
+import java.util.List;
+
+/**
+ * 智谱AI重排序请求DTO
+ *
+ * @author yang
+ * @date 2026-04-19
+ */
+public record ZhipuRerankRequest(
+        String model,
+        String query,
+        List<String> documents,
+        Integer top_n,
+        Boolean return_documents
+) {
+    private static final ObjectMapper OBJECT_MAPPER = new ObjectMapper();
+
+    /**
+     * 创建智谱重排序请求
+     */
+    public static ZhipuRerankRequest create(String modelName, String query,
+                                            List<String> documents, Integer topN,
+                                            Boolean returnDocuments) {
+        return new ZhipuRerankRequest(
+                modelName,
+                query,
+                documents,
+                topN != null ? topN : documents.size(),
+                returnDocuments != null ? returnDocuments : true
+        );
+    }
+
+    /**
+     * 转换为JSON字符串
+     */
+    public String toJson() {
+        try {
+            return OBJECT_MAPPER.writeValueAsString(this);
+        } catch (JsonProcessingException e) {
+            throw new IllegalArgumentException("序列化智谱重排序请求失败", e);
+        }
+    }
+}

ruoyi-modules/ruoyi-chat/src/main/java/org/ruoyi/domain/dto/response/ZhipuRerankResponse.java

@@ -0,0 +1,65 @@
+package org.ruoyi.domain.dto.response;
+
+import com.fasterxml.jackson.annotation.JsonProperty;
+import org.ruoyi.domain.bo.rerank.RerankResult;
+
+import java.util.List;
+import java.util.stream.Collectors;
+
+/**
+ * 智谱AI重排序响应DTO
+ *
+ * @author yang
+ * @date 2026-04-19
+ */
+public record ZhipuRerankResponse(
+        String model,
+        String object,
+        List<ResultItem> results,
+        UsageInfo usage
+) {
+    /**
+     * 单个重排序结果项
+     */
+    public record ResultItem(
+            Integer index,
+            @JsonProperty("relevance_score")
+            Double relevanceScore,
+            String document
+    ) {}
+
+    /**
+     * Token使用信息
+     */
+    public record UsageInfo(
+            @JsonProperty("total_tokens")
+            Integer totalTokens,
+            @JsonProperty("input_tokens")
+            Integer inputTokens,
+            @JsonProperty("output_tokens")
+            Integer outputTokens
+    ) {}
+
+    /**
+     * 转换为通用RerankResult
+     */
+    public RerankResult toRerankResult(int totalDocs, long durationMs) {
+        if (results == null || results.isEmpty()) {
+            return RerankResult.empty();
+        }
+
+        List<RerankResult.RerankDocument> documents = results.stream()
+                .map(item -> RerankResult.RerankDocument.builder()
+                        .index(item.index())
+                        .relevanceScore(item.relevanceScore())
+                        .document(item.document())
+                        .build())
+                .collect(Collectors.toList());
+
+        return RerankResult.builder()
+                .documents(documents)
+                .totalDocuments(totalDocs)
+                .durationMs(durationMs)
+                .build();
+    }
+}

ruoyi-modules/ruoyi-chat/src/main/java/org/ruoyi/domain/entity/knowledge/KnowledgeInfo.java

@@ -78,6 +78,26 @@ public class KnowledgeInfo extends BaseEntity {
      */
     private String embeddingModel;
 
+    /**
+     * 是否启用重排序（0 否 1是）
+     */
+    private Integer enableRerank;
+
+    /**
+     * 重排序模型名称
+     */
+    private String rerankModel;
+
+    /**
+     * 重排序后返回的文档数量
+     */
+    private Integer rerankTopN;
+
+    /**
+     * 重排序相关性分数阈值
+     */
+    private Double rerankScoreThreshold;
+
     /**
      * 备注
      */

ruoyi-modules/ruoyi-chat/src/main/java/org/ruoyi/domain/vo/knowledge/KnowledgeInfoVo.java

@@ -94,6 +94,30 @@ public class KnowledgeInfoVo implements Serializable {
     @ExcelProperty(value = "向量模型")
     private String embeddingModel;
 
+    /**
+     * 是否启用重排序（0 否 1是）
+     */
+    @ExcelProperty(value = "是否启用重排序")
+    private Integer enableRerank;
+
+    /**
+     * 重排序模型名称
+     */
+    @ExcelProperty(value = "重排序模型")
+    private String rerankModel;
+
+    /**
+     * 重排序后返回的文档数量
+     */
+    @ExcelProperty(value = "重排序返回数量")
+    private Integer rerankTopN;
+
+    /**
+     * 重排序相关性分数阈值
+     */
+    @ExcelProperty(value = "重排序分数阈值")
+    private Double rerankScoreThreshold;
+
     /**
      * 备注
      */

ruoyi-modules/ruoyi-chat/src/main/java/org/ruoyi/factory/RerankModelFactory.java

@@ -0,0 +1,106 @@
+package org.ruoyi.factory;
+
+import lombok.RequiredArgsConstructor;
+import lombok.extern.slf4j.Slf4j;
+import org.ruoyi.common.chat.domain.vo.chat.ChatModelVo;
+import org.ruoyi.common.chat.service.chat.IChatModelService;
+import org.ruoyi.service.rerank.RerankModelService;
+import org.springframework.beans.factory.NoSuchBeanDefinitionException;
+import org.springframework.context.ApplicationContext;
+import org.springframework.stereotype.Service;
+
+import java.util.ArrayList;
+import java.util.List;
+import java.util.Map;
+import java.util.concurrent.ConcurrentHashMap;
+
+/**
+ * 重排序模型工厂服务类
+ * 参考设计模式：EmbeddingModelFactory
+ * 负责创建和管理重排序模型实例
+ *
+ * @author yang
+ * @date 2026-04-19
+ */
+@Service
+@RequiredArgsConstructor
+@Slf4j
+public class RerankModelFactory {
+
+    private final ApplicationContext applicationContext;
+
+    private final IChatModelService chatModelService;
+
+    /**
+     * 模型缓存，使用ConcurrentHashMap保证线程安全
+     */
+    private final Map<String, RerankModelService> modelCache = new ConcurrentHashMap<>();
+
+    /**
+     * 创建重排序模型实例
+     * 如果模型已存在于缓存中，则直接返回；否则创建新的实例
+     *
+     * @param rerankModelName 重排序模型名称
+     */
+    public RerankModelService createModel(String rerankModelName) {
+        return modelCache.computeIfAbsent(rerankModelName, name -> {
+            ChatModelVo modelConfig = chatModelService.selectModelByName(rerankModelName);
+
+            if (modelConfig == null) {
+                throw new IllegalArgumentException("未找到重排序模型配置，name=" + name);
+            }
+            return createModelInstance(modelConfig.getProviderCode(), modelConfig);
+        });
+    }
+
+    /**
+     * 刷新模型缓存
+     * 根据给定的模型ID从缓存中移除对应的模型
+     *
+     * @param modelId 模型的唯一标识ID
+     */
+    public void refreshModel(Long modelId) {
+        modelCache.remove(modelId);
+    }
+
+    /**
+     * 获取所有支持模型工厂的列表
+     *
+     * @return 支持的模型工厂名称列表
+     */
+    public List<String> getSupportedFactories() {
+        return new ArrayList<>(applicationContext.getBeansOfType(RerankModelService.class)
+                .keySet());
+    }
+
+    /**
+     * 创建具体的模型实例
+     * 根据提供的工厂名称和配置信息创建并配置模型实例
+     *
+     * @param factory 工厂名称，用于标识模型类型（providerCode）
+     * @param config  模型配置信息
+     * @return RerankModelService 配置好的模型实例
+     * @throws IllegalArgumentException 当无法获取指定的模型实例时抛出
+     */
+    private RerankModelService createModelInstance(String factory, ChatModelVo config) {
+        try {
+            // 优先尝试使用 providerCode + "Rerank" 作为 Bean 名称
+            // 例如：zhipu -> zhipuRerank，jina -> jinaRerank
+            String rerankBeanName = factory + "Rerank";
+            RerankModelService model = applicationContext.getBean(rerankBeanName, RerankModelService.class);
+            model.configure(config);
+            log.info("成功创建重排序模型: factory={}, modelName={}", rerankBeanName, config.getModelName());
+            return model;
+        } catch (NoSuchBeanDefinitionException e) {
+            // 如果找不到，尝试使用原始的 providerCode
+            try {
+                RerankModelService model = applicationContext.getBean(factory, RerankModelService.class);
+                model.configure(config);
+                log.info("成功创建重排序模型: factory={}, modelName={}", factory, config.getModelName());
+                return model;
+            } catch (NoSuchBeanDefinitionException ex) {
+                throw new IllegalArgumentException("获取不到重排序模型: " + factory + " 或 " + factory + "Rerank", ex);
+            }
+        }
+    }
+}

ruoyi-modules/ruoyi-chat/src/main/java/org/ruoyi/service/chat/impl/ChatServiceFacade.java

@@ -54,6 +54,7 @@ import org.ruoyi.service.chat.AbstractChatService;
 import org.ruoyi.service.chat.IChatMessageService;
 import org.ruoyi.service.chat.impl.memory.PersistentChatMemoryStore;
 import org.ruoyi.service.knowledge.IKnowledgeInfoService;
+import org.ruoyi.service.retrieval.KnowledgeRetrievalService;
 import org.ruoyi.service.vector.VectorStoreService;
 import org.springframework.stereotype.Service;
 import org.springframework.web.servlet.mvc.method.annotation.SseEmitter;
@@ -89,6 +90,8 @@ public class ChatServiceFacade implements IChatService {
 
     private final VectorStoreService vectorStoreService;
 
+    private final KnowledgeRetrievalService knowledgeRetrievalService;
+
     private final SseEmitterManager sseEmitterManager;
 
     private final IChatMessageService chatMessageService;
@@ -452,8 +455,8 @@ public class ChatServiceFacade implements IChatService {
             // 构建向量查询参数
             QueryVectorBo queryVectorBo = buildQueryVectorBo(chatRequest, knowledgeInfoVo, chatModel);
 
-            // 获取向量查询结果（知识库内容作为系统上下文，放在历史消息之后）
-            List<String> nearestList = vectorStoreService.getQueryVector(queryVectorBo);
+            // 使用知识库检索服务（支持重排序）
+            List<String> nearestList = knowledgeRetrievalService.retrieveTexts(queryVectorBo);
             for (String prompt : nearestList) {
                 // 知识库内容作为系统上下文添加
                 messages.add(new AiMessage(prompt));
@@ -480,6 +483,13 @@ public class ChatServiceFacade implements IChatService {
         queryVectorBo.setVectorModelName(knowledgeInfoVo.getVectorModel());
         queryVectorBo.setEmbeddingModelName(knowledgeInfoVo.getEmbeddingModel());
         queryVectorBo.setMaxResults(knowledgeInfoVo.getRetrieveLimit());
+
+        // 设置重排序参数
+        queryVectorBo.setEnableRerank(knowledgeInfoVo.getEnableRerank() != null && knowledgeInfoVo.getEnableRerank() == 1);
+        queryVectorBo.setRerankModelName(knowledgeInfoVo.getRerankModel());
+        queryVectorBo.setRerankTopN(knowledgeInfoVo.getRerankTopN());
+        queryVectorBo.setRerankScoreThreshold(knowledgeInfoVo.getRerankScoreThreshold());
+
         return queryVectorBo;
     }
 

ruoyi-modules/ruoyi-chat/src/main/java/org/ruoyi/service/rerank/RerankModelService.java

@@ -0,0 +1,70 @@
+package org.ruoyi.service.rerank;
+
+import dev.langchain4j.data.segment.TextSegment;
+import dev.langchain4j.model.output.Response;
+import dev.langchain4j.model.scoring.ScoringModel;
+import org.ruoyi.common.chat.domain.vo.chat.ChatModelVo;
+import org.ruoyi.domain.bo.rerank.RerankRequest;
+import org.ruoyi.domain.bo.rerank.RerankResult;
+
+import java.util.List;
+
+/**
+ * 重排序模型服务接口
+ * 继承 langchain4j 的 ScoringModel 接口
+ * 参考设计模式：BaseEmbedModelService
+ *
+ * @author Yzm
+ * @date 2026-04-19
+ */
+public interface RerankModelService extends ScoringModel {
+
+    /**
+     * 根据配置信息配置重排序模型
+     *
+     * @param config 包含模型配置信息的 ChatModelVo 对象
+     */
+    void configure(ChatModelVo config);
+
+    /**
+     * 执行重排序（批量文档）
+     * 这是业务层使用的便捷方法
+     *
+     * @param rerankRequest 重排序请求，包含查询文本和候选文档列表
+     * @return 重排序结果，包含排序后的文档和相关性分数
+     */
+    RerankResult rerank(RerankRequest rerankRequest);
+
+    /**
+     * 实现 ScoringModel 接口的 scoreAll 方法
+     * 将 ScoringModel 的调用转换为重排序调用
+     */
+    @Override
+    default Response<List<Double>> scoreAll(List<TextSegment> segments, String query) {
+        // 将 TextSegment 转换为文档字符串列表
+        List<String> documents = segments.stream()
+                .map(TextSegment::text)
+                .toList();
+
+        RerankRequest request = RerankRequest.builder()
+                .query(query)
+                .documents(documents)
+                .topN(documents.size())
+                .returnDocuments(false)
+                .build();
+
+        RerankResult result = rerank(request);
+
+        // 提取分数列表，按原始顺序排列
+        List<Double> scores = new java.util.ArrayList<>(
+                java.util.Collections.nCopies(documents.size(), 0.0));
+
+        for (RerankResult.RerankDocument doc : result.getDocuments()) {
+            if (doc.getIndex() != null && doc.getIndex() < documents.size()) {
+                scores.set(doc.getIndex(), doc.getRelevanceScore());
+            }
+        }
+
+        return Response.from(scores);
+    }
+}

ruoyi-modules/ruoyi-chat/src/main/java/org/ruoyi/service/rerank/impl/JinaRerankModelService.java

@@ -0,0 +1,107 @@
+package org.ruoyi.service.rerank.impl;
+
+import com.fasterxml.jackson.databind.ObjectMapper;
+import dev.langchain4j.http.client.*;
+import lombok.extern.slf4j.Slf4j;
+import org.ruoyi.common.chat.domain.vo.chat.ChatModelVo;
+import org.ruoyi.domain.bo.rerank.RerankRequest;
+import org.ruoyi.domain.bo.rerank.RerankResult;
+import org.ruoyi.service.rerank.RerankModelService;
+import org.springframework.stereotype.Component;
+
+import java.util.ArrayList;
+import java.util.HashMap;
+import java.util.List;
+import java.util.Map;
+
+/**
+ * Jina AI 重排序模型实现
+ * 参考设计模式：OpenAiEmbeddingProvider
+ * 使用 Jina 官方重排序API
+ *
+ * @author yang
+ * @date 2026-04-19
+ */
+@Slf4j
+@Component("jina")
+public class JinaRerankModelService implements RerankModelService {
+
+    protected ChatModelVo chatModelVo;
+    protected HttpClient httpClient;
+    protected final ObjectMapper objectMapper = new ObjectMapper();
+
+    @Override
+    public void configure(ChatModelVo config) {
+        this.chatModelVo = config;
+        HttpClientBuilder httpClientBuilder = HttpClientBuilderLoader.loadHttpClientBuilder();
+        this.httpClient = httpClientBuilder.build();
+    }
+
+    @Override
+    public RerankResult rerank(RerankRequest rerankRequest) {
+        long startTime = System.currentTimeMillis();
+
+        try {
+            // 构建Jina重排序请求体
+            Map<String, Object> requestBody = new HashMap<>();
+            requestBody.put("model", chatModelVo.getModelName());
+            requestBody.put("query", rerankRequest.getQuery());
+            requestBody.put("documents", rerankRequest.getDocuments());
+            requestBody.put("top_n", rerankRequest.getTopN() != null ?
+                    rerankRequest.getTopN() : rerankRequest.getDocuments().size());
+            requestBody.put("return_documents", rerankRequest.getReturnDocuments());
+
+            // 构建HTTP请求
+            HttpRequest httpRequest = HttpRequest.builder()
+                    .url(chatModelVo.getApiHost())
+                    .method(HttpMethod.POST)
+                    .addHeader("Content-Type", "application/json")
+                    .addHeader("Authorization", "Bearer " + chatModelVo.getApiKey())
+                    .body(objectMapper.writeValueAsString(requestBody))
+                    .build();
+
+            // 发送请求
+            SuccessfulHttpResponse httpResponse = httpClient.execute(httpRequest);
+
+            // 解析响应
+            @SuppressWarnings("unchecked")
+            Map<String, Object> response = objectMapper.readValue(httpResponse.body(), Map.class);
+            return parseResponse(response, rerankRequest.getDocuments().size(),
+                    System.currentTimeMillis() - startTime);
+
+        } catch (Exception e) {
+            log.error("Jina重排序失败: {}", e.getMessage(), e);
+            throw new RuntimeException("重排序服务调用失败: " + e.getMessage(), e);
+        }
+    }
+
+    /**
+     * 解析Jina API响应
+     */
+    @SuppressWarnings("unchecked")
+    private RerankResult parseResponse(Map<String, Object> response, int totalDocs, long durationMs) {
+        List<Map<String, Object>> results = (List<Map<String, Object>>) response.get("results");
+        if (results == null || results.isEmpty()) {
+            return RerankResult.empty();
+        }
+
+        List<RerankResult.RerankDocument> documents = new ArrayList<>();
+        for (Map<String, Object> result : results) {
+            Integer index = (Integer) result.get("index");
+            Double score = ((Number) result.get("relevance_score")).doubleValue();
+            String document = (String) result.get("document");
+
+            documents.add(RerankResult.RerankDocument.builder()
+                    .index(index)
+                    .relevanceScore(score)
+                    .document(document)
+                    .build());
+        }
+
+        return RerankResult.builder()
+                .documents(documents)
+                .totalDocuments(totalDocs)
+                .durationMs(durationMs)
+                .build();
+    }
+}

ruoyi-modules/ruoyi-chat/src/main/java/org/ruoyi/service/rerank/impl/ZhiPuRerankModelService.java

@@ -0,0 +1,162 @@
+package org.ruoyi.service.rerank.impl;
+
+import com.fasterxml.jackson.databind.ObjectMapper;
+import io.jsonwebtoken.Jwts;
+import io.jsonwebtoken.security.MacAlgorithm;
+import lombok.extern.slf4j.Slf4j;
+import okhttp3.*;
+import org.ruoyi.common.chat.domain.vo.chat.ChatModelVo;
+import org.ruoyi.domain.bo.rerank.RerankRequest;
+import org.ruoyi.domain.bo.rerank.RerankResult;
+import org.ruoyi.domain.dto.request.ZhipuRerankRequest;
+import org.ruoyi.domain.dto.response.ZhipuRerankResponse;
+import org.ruoyi.service.rerank.RerankModelService;
+import org.springframework.stereotype.Component;
+
+import javax.crypto.spec.SecretKeySpec;
+import java.io.IOException;
+import java.lang.reflect.Constructor;
+import java.nio.charset.StandardCharsets;
+import java.util.concurrent.TimeUnit;
+
+/**
+ * 智谱AI 重排序模型实现
+ * 参考设计模式：AliBaiLianMultiEmbeddingProvider
+ *
+ * @author yang
+ * @date 2026-04-19
+ */
+@Slf4j
+@Component("zhipuRerank")
+public class ZhiPuRerankModelService implements RerankModelService {
+
+    private final OkHttpClient okHttpClient;
+    private final ObjectMapper objectMapper = new ObjectMapper();
+    private ChatModelVo chatModelVo;
+
+    public ZhiPuRerankModelService() {
+        this.okHttpClient = new OkHttpClient.Builder()
+                .connectTimeout(30, TimeUnit.SECONDS)
+                .readTimeout(60, TimeUnit.SECONDS)
+                .writeTimeout(30, TimeUnit.SECONDS)
+                .build();
+    }
+
+    @Override
+    public void configure(ChatModelVo config) {
+        this.chatModelVo = config;
+    }
+
+    @Override
+    public RerankResult rerank(RerankRequest rerankRequest) {
+        long startTime = System.currentTimeMillis();
+
+        try {
+            // 构建请求
+            ZhipuRerankRequest request = buildRequest(rerankRequest);
+            ZhipuRerankResponse response = executeRequest(request);
+
+            return response.toRerankResult(
+                    rerankRequest.getDocuments().size(),
+                    System.currentTimeMillis() - startTime
+            );
+
+        } catch (Exception e) {
+            log.error("智谱重排序失败: {}", e.getMessage(), e);
+            throw new RuntimeException("重排序服务调用失败: " + e.getMessage(), e);
+        }
+    }
+
+    /**
+     * 构建请求对象
+     */
+    private ZhipuRerankRequest buildRequest(RerankRequest rerankRequest) {
+        return ZhipuRerankRequest.create(
+                chatModelVo.getModelName(),
+                rerankRequest.getQuery(),
+                rerankRequest.getDocuments(),
+                rerankRequest.getTopN(),
+                rerankRequest.getReturnDocuments()
+        );
+    }
+
+    /**
+     * 执行HTTP请求并解析响应
+     */
+    private ZhipuRerankResponse executeRequest(ZhipuRerankRequest request) throws IOException {
+        String jsonBody = request.toJson();
+        RequestBody body = RequestBody.create(jsonBody, MediaType.get("application/json"));
+
+        // 生成智谱认证Token
+        String token = generateToken(chatModelVo.getApiKey());
+
+        String url = chatModelVo.getApiHost() + "/" + chatModelVo.getModelName();
+        Request httpRequest = new Request.Builder()
+                .url(url)
+                .addHeader("Authorization", token)
+                .post(body)
+                .build();
+
+        try (okhttp3.Response response = okHttpClient.newCall(httpRequest).execute()) {
+            if (!response.isSuccessful()) {
+                String err = response.body() != null ? response.body().string() : "无错误信息";
+                throw new IllegalArgumentException("智谱API调用失败: " + response.code() + " - " + err);
+            }
+
+            ResponseBody responseBody = response.body();
+            if (responseBody == null) {
+                throw new IllegalArgumentException("响应体为空");
+            }
+
+            return parseResponse(responseBody.string());
+        }
+    }
+
+    /**
+     * 解析响应
+     */
+    private ZhipuRerankResponse parseResponse(String responseBody) throws IOException {
+        return objectMapper.readValue(responseBody, ZhipuRerankResponse.class);
+    }
+
+    /**
+     * 生成智谱JWT Token
+     */
+    private String generateToken(String apiKey) {
+        try {
+            String[] apiKeyParts = apiKey.split("\\.");
+            String keyId = apiKeyParts[0];
+            String secret = apiKeyParts[1];
+
+            long expireMillis = 1000L * 60 * 30; // 30分钟
+            java.util.Map<String, Object> payload = new java.util.HashMap<>();
+            payload.put("api_key", keyId);
+            payload.put("exp", System.currentTimeMillis() + expireMillis);
+            payload.put("timestamp", System.currentTimeMillis());
+
+            // 使用反射创建 MacAlgorithm（兼容不同版本的 jjwt）
+            MacAlgorithm macAlgorithm;
+            try {
+                Class<?> c = Class.forName("io.jsonwebtoken.impl.security.DefaultMacAlgorithm");
+                Constructor<?> ctor = c.getDeclaredConstructor(String.class, String.class, int.class);
+                ctor.setAccessible(true);
+                macAlgorithm = (MacAlgorithm) ctor.newInstance("HS256", "HmacSHA256", 128);
+            } catch (Exception e) {
+                macAlgorithm = Jwts.SIG.HS256;
+            }
+
+            String token = Jwts.builder()
+                    .header()
+                    .add("alg", "HS256")
+                    .add("sign_type", "SIGN")
+                    .and()
+                    .content(objectMapper.writeValueAsString(payload))
+                    .signWith(new SecretKeySpec(secret.getBytes(StandardCharsets.UTF_8), "HmacSHA256"), macAlgorithm)
+                    .compact();
+
+            return "Bearer " + token;
+        } catch (Exception e) {
+            throw new RuntimeException("生成智谱Token失败: " + e.getMessage(), e);
+        }
+    }
+}

ruoyi-modules/ruoyi-chat/src/main/java/org/ruoyi/service/retrieval/KnowledgeRetrievalService.java

@@ -0,0 +1,24 @@
+package org.ruoyi.service.retrieval;
+
+import org.ruoyi.domain.bo.vector.QueryVectorBo;
+
+import java.util.List;
+
+/**
+ * 知识库检索服务接口
+ * 整合粗召回（向量检索）和重排序流程
+ *
+ * @author yang
+ * @date 2026-04-19
+ */
+public interface KnowledgeRetrievalService {
+
+    /**
+     * 执行知识库检索，返回文本内容
+     * 流程：向量粗召回 -> 重排序（可选） -> 返回结果
+     *
+     * @param queryVectorBo 查询参数
+     * @return 文本内容列表
+     */
+    List<String> retrieveTexts(QueryVectorBo queryVectorBo);
+}

ruoyi-modules/ruoyi-chat/src/main/java/org/ruoyi/service/retrieval/impl/KnowledgeRetrievalServiceImpl.java

@@ -0,0 +1,135 @@
+package org.ruoyi.service.retrieval.impl;
+
+import lombok.RequiredArgsConstructor;
+import lombok.extern.slf4j.Slf4j;
+import org.ruoyi.domain.bo.rerank.RerankRequest;
+import org.ruoyi.domain.bo.rerank.RerankResult;
+import org.ruoyi.domain.bo.vector.QueryVectorBo;
+import org.ruoyi.factory.RerankModelFactory;
+import org.ruoyi.service.rerank.RerankModelService;
+import org.ruoyi.service.retrieval.KnowledgeRetrievalService;
+import org.ruoyi.service.vector.VectorStoreService;
+import org.springframework.stereotype.Service;
+
+import java.util.ArrayList;
+import java.util.List;
+
+/**
+ * 知识库检索服务实现
+ * 整合粗召回（向量检索）和重排序流程
+ *
+ * @author yang
+ * @date 2026-04-19
+ */
+@Slf4j
+@Service
+@RequiredArgsConstructor
+public class KnowledgeRetrievalServiceImpl implements KnowledgeRetrievalService {
+
+    private final VectorStoreService vectorStoreService;
+    private final RerankModelFactory rerankModelFactory;
+
+    /**
+     * 粗召回默认扩大倍数
+     * 如果启用重排序，粗召回会获取更多结果供重排序筛选
+     */
+    private static final int RERANK_EXPANSION_FACTOR = 3;
+
+    @Override
+    public List<String> retrieveTexts(QueryVectorBo queryVectorBo) {
+        log.info("开始知识库检索, kid={}, query={}", queryVectorBo.getKid(), queryVectorBo.getQuery());
+
+        // 1. 粗召回阶段 - 向量检索
+        List<String> coarseResults = coarseRetrieval(queryVectorBo);
+        log.debug("粗召回返回 {} 条结果", coarseResults.size());
+
+        if (coarseResults.isEmpty()) {
+            return coarseResults;
+        }
+
+        // 2. 重排序阶段（可选）
+        if (Boolean.TRUE.equals(queryVectorBo.getEnableRerank()) &&
+                queryVectorBo.getRerankModelName() != null) {
+            return rerank(queryVectorBo, coarseResults);
+        }
+
+        return coarseResults;
+    }
+
+    /**
+     * 粗召回阶段 - 向量检索
+     */
+    private List<String> coarseRetrieval(QueryVectorBo queryVectorBo) {
+        // 如果启用重排序，扩大粗召回数量
+        int originalMaxResults = queryVectorBo.getMaxResults();
+        int expandedResults = originalMaxResults;
+        if (Boolean.TRUE.equals(queryVectorBo.getEnableRerank()) &&
+                queryVectorBo.getRerankModelName() != null) {
+            expandedResults = originalMaxResults * RERANK_EXPANSION_FACTOR;
+            log.debug("启用重排序，粗召回数量从 {} 扩大到 {}", originalMaxResults, expandedResults);
+        }
+
+        // 临时修改查询数量
+        queryVectorBo.setMaxResults(expandedResults);
+        try {
+            return vectorStoreService.getQueryVector(queryVectorBo);
+        } finally {
+            // 恢复原始值
+            queryVectorBo.setMaxResults(originalMaxResults);
+        }
+    }
+
+    /**
+     * 重排序阶段
+     */
+    private List<String> rerank(QueryVectorBo queryVectorBo, List<String> coarseResults) {
+        long startTime = System.currentTimeMillis();
+
+        try {
+            // 1. 通过工厂获取重排序模型
+            RerankModelService rerankModel = rerankModelFactory.createModel(queryVectorBo.getRerankModelName());
+
+            // 2. 构建重排序请求
+            int topN = queryVectorBo.getRerankTopN() != null ?
+                    queryVectorBo.getRerankTopN() : queryVectorBo.getMaxResults();
+
+            RerankRequest rerankRequest = RerankRequest.builder()
+                    .query(queryVectorBo.getQuery())
+                    .documents(coarseResults)
+                    .topN(topN)
+                    .returnDocuments(true)
+                    .build();
+
+            log.info("执行重排序, model={}, documents={}, topN={}",
+                    queryVectorBo.getRerankModelName(), coarseResults.size(), topN);
+
+            // 3. 执行重排序
+            RerankResult rerankResult = rerankModel.rerank(rerankRequest);
+
+            // 4. 转换重排序结果
+            List<String> finalResults = new ArrayList<>();
+            for (RerankResult.RerankDocument doc : rerankResult.getDocuments()) {
+                // 应用分数阈值过滤
+                if (queryVectorBo.getRerankScoreThreshold() != null &&
+                        doc.getRelevanceScore() < queryVectorBo.getRerankScoreThreshold()) {
+                    continue;
+                }
+
+                if (doc.getDocument() != null) {
+                    finalResults.add(doc.getDocument());
+                }
+            }
+
+            long duration = System.currentTimeMillis() - startTime;
+            log.info("重排序完成, 返回 {} 条结果, 耗时 {}ms", finalResults.size(), duration);
+
+            return finalResults;
+
+        } catch (Exception e) {
+            log.error("重排序失败: {}", e.getMessage(), e);
+            // 重排序失败时返回原始粗召回结果（截取到期望数量）
+            int limit = Math.min(queryVectorBo.getMaxResults(), coarseResults.size());
+            return new ArrayList<>(coarseResults.subList(0, limit));
+        }
+    }
+}