写入章节大纲

llm-back/pom.xml

@@ -27,7 +27,7 @@
         <fastjson.version>2.0.53</fastjson.version>
         <oshi.version>6.6.5</oshi.version>
         <commons.io.version>2.13.0</commons.io.version>
-        <poi.version>4.1.2</poi.version>
+        <poi.version>5.2.5</poi.version>
         <velocity.version>2.3</velocity.version>
         <jwt.version>0.9.1</jwt.version>
         <!-- override dependency version -->

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

@@ -26,8 +26,8 @@ import io.milvus.param.dml.SearchParam;
 import io.milvus.response.SearchResultsWrapper;
 import org.apache.poi.extractor.POITextExtractor;
 import org.apache.poi.extractor.ExtractorFactory;
-import org.apache.poi.xwpf.extractor.XWPFWordExtractor;
 import org.apache.poi.xwpf.usermodel.*;
+import org.apache.xmlbeans.XmlCursor;
 import org.noear.solon.Solon;
 import org.noear.solon.ai.annotation.ToolMapping;
 import org.noear.solon.ai.chat.ChatModel;
@@ -66,8 +66,9 @@ public class McpServiceImpl implements IMcpService {
     {
             try {
                 title = String.join(",", extractSubTitles( "/upload/agent/template/technical.docx", title));
-                List<JSONObject> contexts = retrieveFromMilvus(milvusClient, embeddingModel, collectionName, title, 10);
-                return generateAnswerWithDocumentAndCollection(embeddingModel, agentName, templatePath, title, contexts, "http://192.168.28.188:8000/v1/chat/completions");
+//                List<JSONObject> contexts = retrieveFromMilvus(milvusClient, embeddingModel, collectionName, title, 10);
+//                return generateAnswerWithDocumentAndCollection(embeddingModel, agentName, templatePath, title, contexts, "http://192.168.28.188:8000/v1/chat/completions");
+                return generateAnswerWithDocumentAndCollection(embeddingModel, agentName, templatePath, title, new ArrayList<>(), "http://192.168.28.188:8000/v1/chat/completions");
             } catch (IOException e) {
                 throw new RuntimeException(e);
             }
@@ -107,9 +108,11 @@ public class McpServiceImpl implements IMcpService {
         Embedding queryEmbedding = embeddingModel.embed(question).content();
         EmbeddingSearchRequest embeddingSearchRequest = EmbeddingSearchRequest.builder()
                 .queryEmbedding(queryEmbedding)
-                .maxResults(3)
+                .minScore(0.7)
                 .build();
-        for (EmbeddingMatch embeddingMatch : embeddingStore.search(embeddingSearchRequest).matches()) {
+        List<EmbeddingMatch<TextSegment>> results = embeddingStore.search(embeddingSearchRequest).matches();
+        results.sort(Comparator.comparingDouble(EmbeddingMatch<TextSegment>::score).reversed());
+        for (EmbeddingMatch<TextSegment> embeddingMatch : results) {
             String requests = embeddingMatch.embedded().toString();
             sb.append(requests).append("\n\n");
 
@@ -156,52 +159,52 @@ public class McpServiceImpl implements IMcpService {
             String[] contentLines = content.split("\n");
             Map<String, String> map = new HashMap<>();
             String[] titles = question.split(",");
+
             for (int i = 0; i < titles.length; i++) {
                 int startIndex = Arrays.asList(contentLines).indexOf(titles[i]);
-                StringBuilder text = new StringBuilder("");
-                if (i < titles.length - 1) {
-                    int endIndex = Arrays.asList(contentLines).indexOf(titles[i + 1]);
-                    for (int c = startIndex + 1; c < endIndex; c++)
-                        text.append(contentLines[c]);
-                } else {
-                    if (startIndex + 1 < contentLines.length)
-                        for (int c = startIndex + 1; c < contentLines.length; c++)
-                            text.append(contentLines[c]);
+                StringBuilder text = new StringBuilder();
+                if (startIndex >= 0) {
+                    if (i < titles.length - 1) {
+                        int endIndex = Arrays.asList(contentLines).indexOf(titles[i + 1]);
+                        for (int c = startIndex + 1; c < endIndex; c++) {
+                            text.append(contentLines[c]).append("\n\n");
+                        }
+                    } else {
+                        if (startIndex + 1 < contentLines.length) {
+                            for (int c = startIndex + 1; c < contentLines.length; c++) {
+                                text.append(contentLines[c]).append("\n\n");
+                            }
+                        }
+                    }
                 }
+                else
+                    text.append(content);
                 map.put(titles[i], text.toString());
             }
-            int count = 0;
-            int position = 0;
+
             List<Integer> positions = new ArrayList<>();
             List<String> contents = new ArrayList<>();
-            for (int i = 0; i < document.getBodyElements().size(); i++) {
-                IBodyElement element = document.getBodyElements().get(i);
-                if (element instanceof XWPFParagraph) {
-                    XWPFParagraph paragraph = (XWPFParagraph) element;
-                    for (String title : titles) {
-                        if (paragraph.getText().equals(title)) {
-                            int pos = position + 1;
-                            positions.add(pos);
-                            contents.add(map.get(title));
-                            count++;
-                        }
+            List<XWPFParagraph> paragraphs = document.getParagraphs();
+
+            for (int i = 0; i < paragraphs.size(); i++) {
+                XWPFParagraph paragraph = paragraphs.get(i);
+                for (String title : titles) {
+                    if (paragraph.getText().contains(title)) {
+                        positions.add(i);
+                        contents.add(map.get(title));
                     }
-                    position++;
-                    if (count == titles.length)
-                        break;
-                }
-                else if (element instanceof XWPFTable) {
-                    XWPFTable table = (XWPFTable) element;
-                    position += table.getNumberOfRows();
                 }
             }
+
             for (int i = positions.size() - 1; i >= 0; i--) {
-                XWPFParagraph contentParagraph = document.createParagraph();
+                int insertPos = positions.get(i) + 1;
+                XmlCursor xmlCursor = paragraphs.get(insertPos).getCTP().newCursor();
+                XWPFParagraph contentParagraph = document.insertNewParagraph(xmlCursor);
                 contentParagraph.setStyle("1");
-                document.setParagraph(contentParagraph, positions.get(i));
                 XWPFRun run = contentParagraph.createRun();
                 run.setText(contents.get(i));
             }
+
             try (FileOutputStream out = new FileOutputStream(absolutePath)) {
                 document.write(out);
             }
@@ -311,7 +314,7 @@ public class McpServiceImpl implements IMcpService {
         else {
             throw new UnsupportedOperationException("不支持文件类型: " + filename);
         }
-        DocumentByParagraphSplitter splitter = new DocumentByParagraphSplitter(1000,200);
+        DocumentByParagraphSplitter splitter = new DocumentByParagraphSplitter(300,50);
         return splitter.split(document);
     }
 

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

@@ -207,9 +207,11 @@ public class LangChainMilvusServiceImpl implements ILangChainMilvusService
             Embedding queryEmbedding = embeddingModel.embed(question).content();
             EmbeddingSearchRequest embeddingSearchRequest = EmbeddingSearchRequest.builder()
                     .queryEmbedding(queryEmbedding)
-                    .maxResults(1)
+                    .minScore(0.7)
                     .build();
-            for (EmbeddingMatch embeddingMatch : embeddingStore.search(embeddingSearchRequest).matches()) {
+            List<EmbeddingMatch<TextSegment>> results = embeddingStore.search(embeddingSearchRequest).matches();
+            results.sort(Comparator.comparingDouble(EmbeddingMatch<TextSegment>::score).reversed());
+            for (EmbeddingMatch<TextSegment> embeddingMatch : results) {
                 String contexts = embeddingMatch.embedded().toString();
                 sb.append("文件").append(": ")
                         .append(document.getPath()).append("\n\n")
@@ -366,7 +368,7 @@ public class LangChainMilvusServiceImpl implements ILangChainMilvusService
         else {
             throw new UnsupportedOperationException("不支持文件类型: " + filename);
         }
-        DocumentByParagraphSplitter splitter = new DocumentByParagraphSplitter(1000,200);
+        DocumentByParagraphSplitter splitter = new DocumentByParagraphSplitter(300,50);
         return splitter.split(document);
     }
 

llm-back/ruoyi-system/pom.xml

@@ -23,6 +23,44 @@
             <artifactId>ruoyi-common</artifactId>
         </dependency>
 
+        <dependency>
+            <groupId>dev.langchain4j</groupId>
+            <artifactId>langchain4j-core</artifactId>
+            <version>0.35.0</version>
+        </dependency>
+
+        <dependency>
+            <groupId>dev.langchain4j</groupId>
+            <artifactId>langchain4j</artifactId>
+            <version>0.35.0</version>
+        </dependency>
+
+        <dependency>
+            <groupId>dev.langchain4j</groupId>
+            <artifactId>langchain4j-document-parser-apache-pdfbox</artifactId>
+            <version>0.35.0</version>
+            <scope>compile</scope>
+        </dependency>
+
+        <dependency>
+            <groupId>dev.langchain4j</groupId>
+            <artifactId>langchain4j-embeddings-bge-small-zh-v15</artifactId>
+            <version>0.35.0</version>
+            <scope>compile</scope>
+        </dependency>
+
+        <dependency>
+            <groupId>org.noear</groupId>
+            <artifactId>solon-ai-mcp</artifactId>
+            <version>3.3.1</version>
+            <exclusions>
+                <exclusion>
+                    <groupId>org.slf4j</groupId>
+                    <artifactId>*</artifactId>
+                </exclusion>
+            </exclusions>
+        </dependency>
+
     </dependencies>
 
 </project>

llm-back/ruoyi-system/src/main/java/com/ruoyi/llm/service/impl/CmcAgentServiceImpl.java

@@ -1,8 +1,7 @@
 package com.ruoyi.llm.service.impl;
 
 import java.io.*;
-import java.util.Date;
-import java.util.List;
+import java.util.*;
 
 import com.alibaba.fastjson2.JSONObject;
 import com.ruoyi.common.config.RuoYiConfig;
@@ -13,7 +12,28 @@ import com.ruoyi.llm.domain.CmcChat;
 import com.ruoyi.llm.domain.CmcDocument;
 import com.ruoyi.llm.mapper.CmcChatMapper;
 import com.ruoyi.llm.mapper.CmcDocumentMapper;
+import dev.langchain4j.data.document.Document;
+import dev.langchain4j.data.document.parser.TextDocumentParser;
+import dev.langchain4j.data.document.parser.apache.pdfbox.ApachePdfBoxDocumentParser;
+import dev.langchain4j.data.document.splitter.DocumentByParagraphSplitter;
+import dev.langchain4j.data.embedding.Embedding;
+import dev.langchain4j.data.segment.TextSegment;
+import dev.langchain4j.model.embedding.EmbeddingModel;
+import dev.langchain4j.model.embedding.onnx.bgesmallzhv15.BgeSmallZhV15EmbeddingModel;
+import dev.langchain4j.store.embedding.EmbeddingMatch;
+import dev.langchain4j.store.embedding.EmbeddingSearchRequest;
+import dev.langchain4j.store.embedding.inmemory.InMemoryEmbeddingStore;
+import org.apache.poi.extractor.ExtractorFactory;
+import org.apache.poi.extractor.POITextExtractor;
+import org.apache.poi.xwpf.usermodel.BreakType;
 import org.apache.poi.xwpf.usermodel.XWPFDocument;
+import org.apache.poi.xwpf.usermodel.XWPFParagraph;
+import org.apache.poi.xwpf.usermodel.XWPFRun;
+import org.noear.solon.ai.chat.ChatModel;
+import org.noear.solon.ai.chat.ChatResponse;
+import org.noear.solon.ai.chat.ChatSession;
+import org.noear.solon.ai.chat.ChatSessionDefault;
+import org.noear.solon.ai.chat.message.ChatMessage;
 import org.springframework.beans.factory.annotation.Autowired;
 import org.springframework.stereotype.Service;
 import com.ruoyi.llm.mapper.CmcAgentMapper;
@@ -39,6 +59,10 @@ public class CmcAgentServiceImpl implements ICmcAgentService
     @Autowired
     private CmcChatMapper cmcChatMapper;
 
+    private static final EmbeddingModel embeddingModel = new BgeSmallZhV15EmbeddingModel();
+
+    private static final String llmServiceUrl = "http://192.168.28.188:8000/v1/chat/completions";
+
     /**
      * 查询智能体
      * 
@@ -85,33 +109,34 @@ public class CmcAgentServiceImpl implements ICmcAgentService
      */
     @Override
     public JSONObject uploadDocument(MultipartFile file, String agentName) throws IOException {
-        File profilePath = new File( RuoYiConfig.getProfile() + "/upload/agent/" + agentName);
+        String prefixPath = "/upload/agent/" + agentName;
+        File profilePath = new File( RuoYiConfig.getProfile() + prefixPath);
         if (!profilePath.exists())
             profilePath.mkdirs();
-        File transferFile = new File(profilePath + "/" + file.getOriginalFilename());
-        if (!transferFile.exists())
-            file.transferTo(transferFile);
         String chatId = new SnowFlake().generateId();
         JSONObject jsonObject = new JSONObject();
         jsonObject.put("chatId", chatId);
-        String[] filenameSplit = file.getOriginalFilename().split("\\.");
-        String outputFilename = "/upload/agent/" + agentName + "/" + file.getOriginalFilename()
-                .replace(filenameSplit[filenameSplit.length - 2], filenameSplit[filenameSplit.length - 2] + "_" + agentName);
-        if (file.getOriginalFilename().endsWith(".doc"))
-            outputFilename = outputFilename.replace(".doc", ".docx");
+        File transferFile = new File(profilePath + "/" + file.getOriginalFilename());
+        if (!transferFile.exists())
+            file.transferTo(transferFile);
         CmcDocument cmcDocument = new CmcDocument();
         cmcDocument.setDocumentId(new SnowFlake().generateId());
         cmcDocument.setChatId(chatId);
-        cmcDocument.setPath(outputFilename);
+        cmcDocument.setPath(prefixPath + "/" + file.getOriginalFilename());
         cmcDocumentMapper.insertCmcDocument(cmcDocument);
         String message = "";
         if (agentName.contains("技术")) {
             CmcChat cmcChat = new CmcChat();
             cmcChat.setChatId(jsonObject.getString("chatId"));
             cmcChat.setInputTime(new Date());
-            cmcChat.setInput("招标文件地址：" + "/upload/agent/" + agentName + "/" + file.getOriginalFilename());
+            cmcChat.setInput("招标文件地址：" + prefixPath + "/" + file.getOriginalFilename());
             cmcChat.setUserId(SecurityUtils.getUserId());
             cmcChatMapper.insertCmcChat(cmcChat);
+            String[] filenameSplit = file.getOriginalFilename().split("\\.");
+            String outputFilename = prefixPath + "/" + file.getOriginalFilename()
+                    .replace(filenameSplit[filenameSplit.length - 2], filenameSplit[filenameSplit.length - 2] + "_" + agentName);
+            if (file.getOriginalFilename().endsWith(".doc"))
+                outputFilename = outputFilename.replace(".doc", ".docx");
             InputStream fileInputStream = new FileInputStream(RuoYiConfig.getProfile() + "/upload/agent/template/technical.docx");
             try (XWPFDocument doc = new XWPFDocument(fileInputStream)) {
                 // 保存文档到本地文件系统
@@ -119,18 +144,12 @@ public class CmcAgentServiceImpl implements ICmcAgentService
                     doc.write(out);
                 }
             }
+            String chapters = generateAnswerWithDocument(embeddingModel, profilePath + "/" + file.getOriginalFilename(),
+                    RuoYiConfig.getProfile() + outputFilename, "工作大纲", llmServiceUrl);
             message = "好的，我已经收到您上传的招标文件，我将给您提供技术文件模板，您可点击进行预览：" +
                     "【<a href='/profile" + outputFilename + "'> 模版 " + "</a>】\n\n" +
-                    "技术文件涉及多个章节：\n" +
-                    "1 项目条件及特性\n" +
-                    "2 项目重点和难点的分析与对策\n" +
-                    "3 项目策划、技术路线、工作方法、流程及措施\n" +
-                    "4 资源配置\n" +
-                    "5 进度计划及保证措施\n" +
-                    "6 质量保证措施\n" +
-                    "7 服务与技术支持\n" +
-                    "8 职业健康、安全生产及环保水保措施\n\n" +
-                    "请问您需要哪些章节撰写的帮助？";
+                    chapters + "\n\n" +
+                    "请问您需要哪个章节撰写的帮助？";
         }
         jsonObject.put("assistantMessage", message);
         return jsonObject;
@@ -203,4 +222,113 @@ public class CmcAgentServiceImpl implements ICmcAgentService
     {
         return cmcAgentMapper.deleteCmcAgentByAgentId(agentId);
     }
+
+
+    /**
+     * 调用LLM生成回答
+     */
+    public String generateAnswerWithDocument(EmbeddingModel embeddingModel, String uploadFilePath, String templatePath, String question, String llmServiceUrl) throws IOException {
+        StringBuilder sb = new StringBuilder("招标文件内容：\n\n");
+        File profilePath = new File(uploadFilePath);
+        List<TextSegment> segments = splitDocument(profilePath);
+        List<Embedding> embeddings = embeddingModel.embedAll(segments).content();
+        InMemoryEmbeddingStore<TextSegment> embeddingStore = new InMemoryEmbeddingStore<>();
+        embeddingStore.addAll(embeddings, segments);
+        Embedding queryEmbedding = embeddingModel.embed(question).content();
+        EmbeddingSearchRequest embeddingSearchRequest = EmbeddingSearchRequest.builder()
+                .queryEmbedding(queryEmbedding)
+                .minScore(0.7)
+                .build();
+        List<EmbeddingMatch<TextSegment>> results = embeddingStore.search(embeddingSearchRequest).matches();
+        results.sort(Comparator.comparingDouble(EmbeddingMatch<TextSegment>::score).reversed());
+        for (EmbeddingMatch<TextSegment> embeddingMatch : results) {
+            String requests = embeddingMatch.embedded().toString();
+            sb.append(requests).append("\n\n");
+        }
+        sb.append("请根据上述招标文件内容，严格按以下格式列出").append(question).append(":\n")
+                .append("6.1 XX\n" +
+                "6.2 XX\n" +
+                "6.3 XX");
+        return generateAnswer(sb.toString(), question, templatePath, llmServiceUrl);
+    }
+
+    /**
+     * 调用LLM生成回答
+     * @return
+     */
+    public String generateAnswer(String prompt, String question, String templatePath, String llmServiceUrl) throws IOException {
+        ChatSession chatSession = new ChatSessionDefault();
+        ChatModel chatModel = ChatModel.of(llmServiceUrl)
+                .provider("openai")
+                .model("Qwen2.5-1.5B-Instruct")
+                .build();
+
+        chatSession.addMessage(ChatMessage.ofUser(prompt));
+
+        ChatResponse response = chatModel.prompt(chatSession).call();
+        String content = response.lastChoice().getMessage().getContent();
+        writeContent(content, templatePath);
+        return content;
+    }
+
+    /**
+     * 写入章节大纲
+     * @return
+     */
+    public void writeContent(String content, String templatePath) throws IOException {
+        List<String> chapters = new ArrayList<>();
+        String[] contentLines = content.split("\n");
+        for (String line : contentLines) {
+            if (line.contains("6."))
+                chapters.add(line.replace("*", "").replace("#", "").split(" ")[1]);
+        }
+        File file = new File(templatePath);
+        FileInputStream fileInputStream = new FileInputStream(file);
+        try (XWPFDocument document = new XWPFDocument(fileInputStream)) {
+            for (int i = 0; i < chapters.size(); i++) {
+                XWPFParagraph contentParagraph = document.createParagraph();
+                contentParagraph.setStyle("3");
+                XWPFRun run = contentParagraph.createRun();
+                run.setText(chapters.get(i));
+                run.addBreak();
+                if (i < chapters.size() - 1)
+                    run.addBreak(BreakType.PAGE);
+            }
+
+            try (FileOutputStream out = new FileOutputStream(templatePath)) {
+                document.write(out);
+            }
+        }
+    }
+
+    /**
+     * 检索知识库
+     */
+    private List<TextSegment> splitDocument(File transferFile) throws IOException {
+        // 加载文档
+        Document document;
+        InputStream fileInputStream = new FileInputStream(transferFile);
+        String filename = transferFile.getName().toLowerCase();
+        if (filename.endsWith(".doc") || filename.endsWith(".docx")) {
+            try (POITextExtractor extractor = ExtractorFactory.createExtractor(fileInputStream)) {
+                String text = extractor.getText();
+                document = Document.from(text);
+            }
+            catch (IOException e) {
+                throw new RuntimeException(e);
+            }
+        }
+        else if (filename.endsWith(".pdf")) {
+            document = new ApachePdfBoxDocumentParser().parse(fileInputStream);
+        }
+        else if (filename.endsWith(".txt")) {
+            document = new TextDocumentParser().parse(fileInputStream);
+        }
+        else {
+            throw new UnsupportedOperationException("不支持文件类型: " + filename);
+        }
+        DocumentByParagraphSplitter splitter = new DocumentByParagraphSplitter(300,50);
+        return splitter.split(document);
+    }
+
 }