去除相关度小于0.7的上下文参考

llm-back/ruoyi-agent/pom.xml

@@ -50,6 +50,11 @@
             </exclusions>
         </dependency>
 
+        <dependency>
+            <groupId>org.apache.poi</groupId>
+            <artifactId>poi-ooxml</artifactId>
+        </dependency>
+
     </dependencies>
 
     <dependencyManagement>

llm-back/ruoyi-agent/src/main/java/com/ruoyi/agent/service/impl/McpServiceImpl.java

@@ -14,7 +14,6 @@ import org.springframework.stereotype.Service;
 import java.io.FileInputStream;
 import java.io.FileOutputStream;
 import java.io.IOException;
-import java.util.Arrays;
 import java.util.Collection;
 import java.util.Collections;
 
@@ -41,7 +40,7 @@ public class McpServiceImpl implements IMcpService {
     @ToolMapping(description = "补全技术方案")
     public String writeTechnicalPlan(@Param(description = "文件路径") String document, @Param(description = "技术方案内容") String resultContent) throws IOException {
         // 创建临时输出文件路径
-        String outputPath = document.replace(".docx", "_modified.docx");
+        String outputPath = document.replace(".docx", "_修改" + System.currentTimeMillis() + ".docx");
 
         try (FileInputStream fis = new FileInputStream(document);
              XWPFDocument doc = new XWPFDocument(fis);

llm-back/ruoyi-llm/src/main/java/com/ruoyi/web/llm/controller/McpController.java

@@ -1,20 +1,19 @@
 package com.ruoyi.web.llm.controller;
 
+import com.alibaba.fastjson2.JSONObject;
+import com.ruoyi.common.config.RuoYiConfig;
 import com.ruoyi.common.core.controller.BaseController;
-import com.ruoyi.web.llm.service.ILangChainMilvusService;
 import org.noear.solon.ai.chat.ChatModel;
 import org.noear.solon.ai.chat.ChatResponse;
 import org.noear.solon.ai.chat.message.AssistantMessage;
 import org.noear.solon.ai.chat.message.ChatMessage;
-import org.noear.solon.ai.chat.tool.FunctionToolDesc;
 import org.noear.solon.ai.mcp.client.McpClientProvider;
-import org.reactivestreams.Publisher;
-import org.springframework.beans.factory.annotation.Autowired;
 import org.springframework.web.bind.annotation.GetMapping;
 import org.springframework.web.bind.annotation.RequestMapping;
 import org.springframework.web.bind.annotation.RestController;
-import reactor.core.publisher.Flux;
+import org.springframework.web.multipart.MultipartFile;
 
+import java.io.File;
 import java.io.IOException;
 import java.util.HashMap;
 import java.util.List;
@@ -31,31 +30,34 @@ import java.util.Map;
 @RequestMapping("/llm/mcp")
 public class McpController extends BaseController
 {
-    @Autowired
-    private ILangChainMilvusService langChainMilvusService;
-
     /**
-     * 打开word文档
+     * 同步问答
      * @return
      */
     @GetMapping("/answer")
-    public Flux<AssistantMessage> answer(String question) throws IOException {
+    public AssistantMessage answer(String question) throws IOException {
         McpClientProvider clientProvider = McpClientProvider.builder()
-                    .apiUrl("http://localhost:8080/llm/mcp/sse")
-                    .build();
+                .apiUrl("http://localhost:8080/llm/mcp/sse")
+                .build();
 
         ChatModel chatModel = ChatModel.of("http://192.168.28.188:8000/v1/chat/completions")
                 .provider("openai")
-                .model("DeepSeek-R1-Distill-Qwen-1.5B")
+                .model("Qwen2.5-1.5B-Instruct")
                 .apiKey("1")
                 .defaultToolsAdd(clientProvider)
                 .build();
-        Publisher<ChatResponse> publisher = chatModel.prompt(question).stream();
-
-        return Flux.from(publisher)
-                .map(response -> {
-                    return response.getChoices().get(0).getMessage();
-                });
+        ChatResponse response = chatModel.prompt(question).call();
+        String resultContent = response.lastChoice().getMessage().getResultContent();
+        AssistantMessage assistantMessage = new AssistantMessage(resultContent);
+        if (resultContent.startsWith("<tool_call>")) {
+            String content = resultContent.replace("<tool_call>\n", "").replace("\n</tool_call>", "");
+            JSONObject jsonObject = JSONObject.parseObject(content);
+            String name = jsonObject.getString("name");
+            JSONObject arguments = jsonObject.getJSONObject("arguments");
+            resultContent = clientProvider.callToolAsText(name, arguments).getContent();
+            assistantMessage = new AssistantMessage(resultContent);
+        }
+        return assistantMessage;
     }
 
     /**
@@ -63,7 +65,7 @@ public class McpController extends BaseController
      * @return
      */
     @GetMapping("/tech")
-    public AssistantMessage tech(String file) throws IOException {
+    public AssistantMessage tech(MultipartFile file) throws IOException {
         McpClientProvider clientProvider = McpClientProvider.builder()
                     .apiUrl("http://localhost:8080/llm/mcp/sse")
                     .build();
@@ -73,19 +75,25 @@ public class McpController extends BaseController
                 .apiKey("1")
                 .defaultToolsAdd(clientProvider)
                 .build();
-
+        File profilePath = new File( RuoYiConfig.getProfile() + "/upload/knowledge");
+        if (!profilePath.exists())
+            profilePath.mkdirs();
+        File transferFile = new File( profilePath + File.separator + file.getOriginalFilename());
+        if (!transferFile.exists()) {
+            file.transferTo(transferFile);
+        }
         Map<String,Object> path = new HashMap<>();
-        path.put("document", file);
+        path.put("document", transferFile.getPath());
         String content = clientProvider.callToolAsText("openDocument", path).getContent();
         Map<String,Object> request = new HashMap<>();
         request.put("request", content);
         List<ChatMessage> messages = clientProvider.getPromptAsMessages("askQuestion", request);
         ChatResponse response = chatModel.prompt(messages).call();
         System.out.println(response.getChoices().get(0).getMessage());
-        Map<String,Object> write = new HashMap<>();
-        write.put("document", file);
-        write.put("resultContent", response.getChoices().get(0).getMessage().getResultContent());
-        clientProvider.callToolAsText("writeTechnicalPlan", write);
+//        Map<String,Object> write = new HashMap<>();
+//        write.put("document", transferFile.getPath());
+//        write.put("resultContent", response.getChoices().get(0).getMessage().getResultContent());
+//        String res = clientProvider.callToolAsText("writeTechnicalPlan", write).getContent();
         return response.getChoices().get(0).getMessage();
     }
 

llm-back/ruoyi-llm/src/main/java/com/ruoyi/web/llm/controller/RagController.java

@@ -1,14 +1,8 @@
-/*
- * @Author: ysh
- * @Date: 2025-07-08 15:10:42
- * @LastEditors: 
- * @LastEditTime: 2025-07-09 11:37:02
- */
 package com.ruoyi.web.llm.controller;
 
+import com.alibaba.fastjson2.JSONObject;
 import com.ruoyi.web.llm.service.ILangChainMilvusService;
 import com.ruoyi.common.core.controller.BaseController;
-import com.ruoyi.common.core.domain.AjaxResult;
 import dev.langchain4j.model.embedding.EmbeddingModel;
 import dev.langchain4j.model.embedding.onnx.bgesmallzhv15.BgeSmallZhV15EmbeddingModel;
 import io.milvus.client.MilvusServiceClient;
@@ -20,7 +14,6 @@ import org.springframework.web.bind.annotation.RequestMapping;
 import org.springframework.web.bind.annotation.RestController;
 import reactor.core.publisher.Flux;
 
-import java.io.IOException;
 import java.util.List;
 
 /**
@@ -48,10 +41,19 @@ public class RagController extends BaseController
      * 增强检索生成回答
      */
     @GetMapping("/answer")
-    public Flux<AssistantMessage> answer(String question, String collectionName) throws IOException {
-
-        List<String> contexts = langChainMilvusService.retrieveFromMilvus(milvusClient, embeddingModel, collectionName, question, 1);
+    public Flux<AssistantMessage> answer(String question, String collectionName)
+    {
+        List<JSONObject> contexts = langChainMilvusService.retrieveFromMilvus(milvusClient, embeddingModel, collectionName, question, 10);
         return langChainMilvusService.generateAnswerWithRag(question, contexts, "http://192.168.28.188:8000/v1/chat/completions");
     }
 
+    /**
+     * 增强检索生成上下文
+     */
+    @GetMapping("/context")
+    public List<JSONObject> context(String question, String collectionName)
+    {
+        return langChainMilvusService.similarityFromMilvus(milvusClient, embeddingModel, collectionName, question, 10);
+    }
+
 }

llm-back/ruoyi-llm/src/main/java/com/ruoyi/web/llm/controller/SessionController.java

@@ -1,7 +1,6 @@
 package com.ruoyi.web.llm.controller;
 
 import com.ruoyi.common.core.controller.BaseController;
-import com.ruoyi.common.core.domain.AjaxResult;
 import com.ruoyi.web.llm.service.ILangChainMilvusService;
 import org.noear.solon.ai.chat.message.AssistantMessage;
 import org.springframework.beans.factory.annotation.Autowired;
@@ -10,8 +9,6 @@ import org.springframework.web.bind.annotation.RequestMapping;
 import org.springframework.web.bind.annotation.RestController;
 import reactor.core.publisher.Flux;
 
-import java.io.IOException;
-
 /**
  * session对话Controller
  * 

llm-back/ruoyi-llm/src/main/java/com/ruoyi/web/llm/service/ILangChainMilvusService.java

@@ -1,5 +1,6 @@
 package com.ruoyi.web.llm.service;
 
+import com.alibaba.fastjson2.JSONObject;
 import dev.langchain4j.model.embedding.EmbeddingModel;
 import io.milvus.client.MilvusClient;
 import io.milvus.grpc.MutationResult;
@@ -20,14 +21,21 @@ public interface ILangChainMilvusService {
 
     /**
      * 从Milvus检索相关文档
+     * @return
+     */
+    public List<JSONObject> retrieveFromMilvus(MilvusClient milvusClient, EmbeddingModel embeddingModel, String collectionName, String query, int topK);
+
+    /**
+     * 从Milvus检索相关文档及相关度
+     * @return
      */
-    public List<String> retrieveFromMilvus(MilvusClient milvusClient, EmbeddingModel embeddingModel, String collectionName, String query, int topK);
+    public List<JSONObject> similarityFromMilvus(MilvusClient milvusClient, EmbeddingModel embeddingModel, String collectionName, String query, int topK);
 
     /**
      * 调用LLM+RAG生成回答
      * @return
      */
-    public Flux<AssistantMessage> generateAnswerWithRag(String question, List<String> contexts, String llmServiceUrl);
+    public Flux<AssistantMessage> generateAnswerWithRag(String question, List<JSONObject> contexts, String llmServiceUrl);
 
     /**
      * 调用LLM生成回答

llm-back/ruoyi-llm/src/main/java/com/ruoyi/web/llm/service/impl/LangChainMilvusServiceImpl.java

@@ -6,6 +6,7 @@
  */
 package com.ruoyi.web.llm.service.impl;
 
+import com.alibaba.fastjson2.JSONObject;
 import com.ruoyi.common.config.RuoYiConfig;
 import com.ruoyi.web.llm.service.ILangChainMilvusService;
 import dev.langchain4j.data.document.Document;
@@ -74,7 +75,7 @@ public class LangChainMilvusServiceImpl implements ILangChainMilvusService
         else {
             throw new UnsupportedOperationException("不支持文件类型: " + filename);
         }
-        DocumentByParagraphSplitter splitter = new DocumentByParagraphSplitter(1024,0);;
+        DocumentByParagraphSplitter splitter = new DocumentByParagraphSplitter(1024,100);;
         List<TextSegment> segments = splitter.split(document);
 
         // 提取文本和生成嵌入
@@ -112,9 +113,80 @@ public class LangChainMilvusServiceImpl implements ILangChainMilvusService
 
     /**
      * 从Milvus检索相关文档
+     * @return
+     */
+    @Override
+    public List<JSONObject> retrieveFromMilvus(MilvusClient milvusClient, EmbeddingModel embeddingModel, String collectionName, String query, int topK) {
+        SearchResultsWrapper wrapper = retrieve(milvusClient, embeddingModel, collectionName, query, topK);
+        return wrapper.getRowRecords(0).stream()
+                .map(record -> {
+                    JSONObject result = new JSONObject();
+                    result.put("file_name", record.get("file_name"));
+                    result.put("content", record.get("content"));
+                    return result;
+                })
+                .collect(Collectors.toList());
+    }
+
+    /**
+     * 从Milvus检索相关文档及相关度
+     * @return
+     */
+    @Override
+    public List<JSONObject> similarityFromMilvus(MilvusClient milvusClient, EmbeddingModel embeddingModel, String collectionName, String query, int topK) {
+        SearchResultsWrapper wrapper = retrieve(milvusClient, embeddingModel, collectionName, query, topK);
+        List<JSONObject> wrapperList =  wrapper.getRowRecords(0).stream()
+                .map(record -> {
+                    JSONObject result = new JSONObject();
+                    result.put("distance", record.get("distance"));
+                    result.put("file_name", record.get("file_name"));
+                    result.put("content", record.get("content"));
+                    return result;
+                })
+                .collect(Collectors.toList());
+        wrapperList.removeIf(jsonObject -> jsonObject.getDouble("distance") < 0.7);
+        return wrapperList;
+    }
+
+    /**
+     * 调用LLM+RAG生成回答
+     * @return
+     */
+    @Override
+    public Flux<AssistantMessage> generateAnswerWithRag(String question, List<JSONObject> contexts, String llmServiceUrl) {
+        StringBuilder sb = new StringBuilder();
+        sb.append("问题: ").append(question).append("\n\n");
+        sb.append("根据以下文件中的上下文内容回答问题：\n\n");
+        for (int i = 0; i < contexts.size(); i++) {
+            sb.append("文件").append(i + 1).append(": ")
+                    .append(contexts.get(i).getString("file_name")).append("\n\n")
+                    .append("上下文").append(": ")
+                    .append(contexts.get(i).getString("content")).append("\n\n");
+        }
+        // 构建带动态参数的URL
+        return generateAnswer(sb.toString(), llmServiceUrl);
+    }
+
+    /**
+     * 调用LLM生成回答
+     * @return
      */
     @Override
-    public List<String> retrieveFromMilvus(MilvusClient milvusClient, EmbeddingModel embeddingModel, String collectionName, String query, int topK) {
+    public Flux<AssistantMessage> generateAnswer(String prompt, String llmServiceUrl) {
+        ChatModel chatModel = ChatModel.of(llmServiceUrl)
+                .provider("openai")
+                .model("Qwen2.5-1.5B-Instruct")
+                .apiKey("1")
+                .build();
+        Publisher<ChatResponse> publisher = chatModel.prompt(prompt).stream();
+
+        return Flux.from(publisher)
+                .map(response -> {
+                    return response.lastChoice().getMessage();
+                });
+    }
+
+    private SearchResultsWrapper retrieve(MilvusClient milvusClient, EmbeddingModel embeddingModel, String collectionName, String query, int topK) {
         List<List<Float>> queryVector = Collections.singletonList(embeddingModel.embed(query).content().vectorAsList());
 
         //  加载集合
@@ -148,44 +220,6 @@ public class LangChainMilvusServiceImpl implements ILangChainMilvusService
                 .build();
         milvusClient.releaseCollection(param);
 
-        return wrapper.getRowRecords(0).stream()
-                .map(record -> (String) record.get("content"))
-                .collect(Collectors.toList());
-    }
-
-    /**
-     * 调用LLM+RAG生成回答
-     * @return
-     */
-    @Override
-    public Flux<AssistantMessage> generateAnswerWithRag(String question, List<String> contexts, String llmServiceUrl) {
-        StringBuilder sb = new StringBuilder();
-        sb.append("根据以下上下文回答问题：\n\n");
-        for (int i = 0; i < contexts.size(); i++) {
-            sb.append("上下文").append(i+1).append(": ").append(contexts.get(i)).append("\n\n");
-        }
-        sb.append("问题: ").append(question).append("\n回答: ");
-        // 构建带动态参数的URL
-        return generateAnswer(sb.toString(),llmServiceUrl);
+        return wrapper;
     }
-
-    /**
-     * 调用LLM生成回答
-     * @return
-     */
-    @Override
-    public Flux<AssistantMessage> generateAnswer(String prompt, String llmServiceUrl) {
-        ChatModel chatModel = ChatModel.of(llmServiceUrl)
-                .provider("openai")
-                .model("DeepSeek-R1-Distill-Qwen-1.5B")
-                .apiKey("1")
-                .build();
-        Publisher<ChatResponse> publisher = chatModel.prompt(prompt).stream();
-
-        return Flux.from(publisher)
-                .map(response -> {
-                    return response.getChoices().get(0).getMessage();
-                });
-    }
-
 }