refactor: ♻️ 使用 Re-reading 对回答进行增强，提高生成内容的质量

Author: StephenQiu30

algorithm-service/algorithm-ai-service/src/main/java/com/stephen/cloud/ai/advisor/ReReadingAdvisor.java

@@ -0,0 +1,59 @@
+package com.stephen.cloud.ai.advisor;
+
+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.prompt.PromptTemplate;
+import org.springframework.stereotype.Component;
+
+import java.util.Map;
+
+@Component
+public class ReReadingAdvisor implements BaseAdvisor {
+
+    private static final String DEFAULT_RE2_ADVISE_TEMPLATE = """
+            {re2_input_query}
+            Read the question again: {re2_input_query}
+            """;
+
+    private final String re2AdviseTemplate;
+
+    private int order = 0;
+
+    public ReReadingAdvisor() {
+        this(DEFAULT_RE2_ADVISE_TEMPLATE);
+    }
+
+    public ReReadingAdvisor(String re2AdviseTemplate) {
+        this.re2AdviseTemplate = re2AdviseTemplate;
+    }
+
+    @Override
+    public ChatClientRequest before(ChatClientRequest chatClientRequest, AdvisorChain advisorChain) {
+        String augmentedUserText = PromptTemplate.builder()
+                .template(this.re2AdviseTemplate)
+                .variables(Map.of("re2_input_query", chatClientRequest.prompt().getUserMessage().getText()))
+                .build()
+                .render();
+
+        return chatClientRequest.mutate()
+                .prompt(chatClientRequest.prompt().augmentUserMessage(augmentedUserText))
+                .build();
+    }
+
+    @Override
+    public ChatClientResponse after(ChatClientResponse chatClientResponse, AdvisorChain advisorChain) {
+        return chatClientResponse;
+    }
+
+    @Override
+    public int getOrder() {
+        return this.order;
+    }
+
+    public ReReadingAdvisor withOrder(int order) {
+        this.order = order;
+        return this;
+    }
+}

algorithm-service/algorithm-ai-service/src/main/java/com/stephen/cloud/ai/config/RagGenerationProperties.java

@@ -17,5 +17,5 @@ public class RagGenerationProperties {
     /**
      * 回答生成温度
      */
-    private Double temperature = 0.7D;
+    private Double temperature = 0.4D;
 }

algorithm-service/algorithm-ai-service/src/main/java/com/stephen/cloud/ai/service/impl/RAGServiceImpl.java

@@ -5,6 +5,7 @@
 import cn.hutool.json.JSONUtil;
 import com.baomidou.mybatisplus.core.conditions.query.LambdaQueryWrapper;
 import com.baomidou.mybatisplus.extension.plugins.pagination.Page;
+import com.stephen.cloud.ai.advisor.ReReadingAdvisor;
 import com.stephen.cloud.ai.config.RagGenerationProperties;
 import com.stephen.cloud.ai.convert.RAGConvert;
 import com.stephen.cloud.ai.knowledge.retrieval.RagDocumentHelper;
@@ -34,6 +35,7 @@
 import org.springframework.ai.chat.messages.Message;
 import org.springframework.ai.chat.messages.MessageType;
 import org.springframework.ai.chat.messages.UserMessage;
+import org.springframework.ai.chat.prompt.PromptTemplate;
 import org.springframework.ai.document.Document;
 import org.springframework.ai.rag.advisor.RetrievalAugmentationAdvisor;
 import org.springframework.ai.rag.generation.augmentation.ContextualQueryAugmenter;
@@ -60,11 +62,24 @@ public class RAGServiceImpl implements RAGService {
 
     private static final String RAG_SYSTEM_PROMPT = """
             你是一个严谨的知识库问答助手。
-            请优先依据检索到的知识片段回答用户问题。
-            如果上下文不足，请明确说明“当前知识库中没有足够信息支持该结论”，不要编造。
-            回答时尽量简洁、准确，并在合适时结合引用片段给出结论。
+            请仅依据检索到的知识片段回答用户问题；不得编造片段中不存在的事实或数据。
+            如果上下文不足，请明确说明“当前知识库中没有足够信息支持该结论”。
+            引用依据时请使用片段序号或片段头中的 chunkId；回答尽量简洁、准确。
             """;
 
+    private static final PromptTemplate RAG_CONTEXTUAL_QUERY_AUGMENT_TEMPLATE = new PromptTemplate("""
+            以下是仅用于本题作答的检索片段：
+
+            ---------------------
+            {context}
+            ---------------------
+
+            用户问题（若含重复表述以辅助阅读，请抓住实质问题）：
+            {query}
+
+            请仅依据上述片段作答；引用时标明片段序号或 chunkId；信息不足时直接说明，不要臆测。
+            """);
+
     private static final String NO_CONTEXT_ANSWER = "当前知识库没有检索到足够相关的内容，暂时无法给出可靠回答。";
 
     @Resource
@@ -85,6 +100,9 @@ public class RAGServiceImpl implements RAGService {
     @Resource
     private RagDocumentHelper ragDocumentHelper;
 
+    @Resource
+    private ReReadingAdvisor reReadingAdvisor;
+
     @Override
     public Flux<String> askStream(String question, Long knowledgeBaseId, Long userId, Integer topK,
             String conversationId) {
@@ -427,16 +445,18 @@ private ChatClient.ChatClientRequestSpec buildRagRequest(String question, String
         RetrievalAugmentationAdvisor ragAdvisor = RetrievalAugmentationAdvisor.builder()
                 .documentRetriever(query -> docs)
                 .queryAugmenter(ContextualQueryAugmenter.builder()
+                        .promptTemplate(RAG_CONTEXTUAL_QUERY_AUGMENT_TEMPLATE)
                         .allowEmptyContext(true)
                         .documentFormatter(this::formatDocumentsForPrompt)
                         .build())
+                .order(1)
                 .build();
 
         MessageChatMemoryAdvisor memoryAdvisor = MessageChatMemoryAdvisor.builder(chatMemory).build();
 
         return chatClient.prompt()
                 .options(buildGenerationOptions())
-                .advisors(memoryAdvisor, ragAdvisor)
+                .advisors(memoryAdvisor, reReadingAdvisor, ragAdvisor)
                 .advisors(advisorSpec -> advisorSpec.param(ChatMemory.CONVERSATION_ID, conversationId))
                 .system(RAG_SYSTEM_PROMPT)
                 .user(question);

nacos-config/common-ai-prod.yml

@@ -40,7 +40,7 @@ rag:
     complex-query-top-k: 15
   generation:
     max-context-length: 6000
-    temperature: 0.5
+    temperature: 0.4
 
 # Chat Memory 配置
 chat:

nacos-config/common-ai.yml

@@ -35,7 +35,7 @@ rag:
     complex-query-top-k: 12            # 复杂问题最少召回片段数
   generation:
     max-context-length: 4000           # 最大上下文长度
-    temperature: 0.7                   # 生成温度
+    temperature: 0.4                   # 生成温度
 
 # Chat Memory 配置
 chat: