[前端+后端+RAG] 检索范围切换(当前文件/整个知识库)；联网搜索功能(SearXNG)；搜索结果带网络链接；修复RAG检索source格式不匹配bug

langchain-chat/server/chat/self_kb_chat.py

@@ -50,6 +50,7 @@ async def self_kb_chat(
                             "content": "虎头虎脑"}]]
                     ),
                     stream: bool = Body(True, description="流式输出"),
+                    web_search: bool = Body(False, description="是否开启联网搜索"),
                     ):
     """
         个人知识库对话api\n
@@ -149,7 +150,29 @@ async def self_kb_chat(
         except Exception as e:
             logger.error(f"个人知识库问答路由错误: {self_kb_route}", exc_info=True)
             docs = []
-        logger.info(f"个人知识库问答source_documents: {docs}")
+        logger.info(f"个人知识库问答source_documents: {len(docs)}条")
+
+        # 联网搜索
+        web_search_context = ""
+        web_search_results = []  # 保存搜索结果供后面引用
+        if web_search:
+            try:
+                from server.chat.ZhipuSearchAPI import ZhipuSearchAPIWrapper
+                searcher = ZhipuSearchAPIWrapper()
+                web_results = searcher.zhipu_search(search_query)
+                web_search_results = web_results[:5] if web_results else []
+                if web_results:
+                    web_parts = []
+                    for i, r in enumerate(web_results[:5], 1):
+                        title = r.get("title", "")
+                        content = r.get("content", "")[:300]
+                        url = r.get("url", "")
+                        web_parts.append(f"[{i}] {title}\n{content}\n来源: {url}")
+                    web_search_context = "\n\n【联网搜索结果】\n" + "\n\n".join(web_parts)
+                    logger.info(f"联网搜索获取到 {len(web_results)} 条结果")
+            except Exception as e:
+                logger.error(f"联网搜索失败: {e}")
+
         # if SELF_USE_RERANKER:
         #     reranker_model_path = MODEL_PATH["reranker"].get(RERANKER_MODEL,"BAAI/bge-reranker-large")
         #     print("-----------------model path------------------")
@@ -184,25 +207,28 @@ async def self_kb_chat(
         if '0' in self_kb_route:
             context = "\n".join([doc.page_content for doc in docs]).strip("xa0")
             logger.info(f"个人知识库问答 context 长度：{len(context)}")
-            # context_70 = context if len(context)<30000 else TextRank(context,num_sentences=70)
-            context = context[:40000] if len(context)>40000 else context
-            logger.info(f"截取后个人知识库问答 context 长度：{len(context)}")
+            context = context[:30000] if len(context)>30000 else context
+            if web_search_context:
+                context += web_search_context
+            logger.info(f"最终 context 长度：{len(context)}")
             if history:
                 history = history if len(history) < 20000 else TextRank(history,num_sentences=1)
-            # logger.info(f"个人知识库问答 context 长度超过 30000，使用 TextRank 算法进行降维得到 context 长度：{len(context)}")
             chain = LLMChain(prompt=chat_prompt, llm=model1, verbose=True)
             task = asyncio.create_task(wrap_done(
                 chain.acall({"context": context, "question": query, "history": history, "quote": quote, "fileName":fileNames}),
                 callback.done),
             )
         elif '1' in self_kb_route:
+            # 联网搜索结果作为额外文档加入
+            if web_search_context:
+                from langchain.docstore.document import Document as LCDocument
+                docs.append(LCDocument(page_content=web_search_context, metadata={"source": "web_search"}))
             chain = load_qa_chain(
-                model, 
-                chain_type="stuff", 
-                prompt=chat_prompt, 
+                model,
+                chain_type="stuff",
+                prompt=chat_prompt,
                 verbose=True
             )
-            # Begin a task that runs in the background.
             task = asyncio.create_task(wrap_done(
                 chain.acall({"input_documents": docs, "question": query, "history": history, "quote": quote, "fileName":fileNames}),
                 callback.done),
@@ -235,14 +261,18 @@ async def self_kb_chat(
             yield json.dumps(response, ensure_ascii=False)
         await task
         source_documents = []
-        if len(docs) == 0:  # 没有找到相关文档
+        if len(docs) == 0 and not web_search_context:
             source_documents.append(f"""暂未从本篇文献中找到答案,该回答为大模型自身能力解答！""")
         else:
-            # 去除文件扩展名
-            # fileNames_without_ext = [name.rsplit('.', 1)[0] for name in fileNames]
-            # 连接文件名（如果有多个文件名）
-            # joined_fileNames = ', '.join(fileNames_without_ext)
-            source_documents.append(f"""[{len(source_documents) + 1}] [{docs[0].metadata.get("source")}]()\n""")    
+            if len(docs) > 0:
+                source_documents.append(f"""[{len(source_documents) + 1}] [{docs[0].metadata.get("source")}]()\n""")
+            # 联网搜索结果链接
+            if web_search_results:
+                for r in web_search_results:
+                    title = r.get("title", "").replace("\n", "")
+                    url = r.get("url", "")
+                    if title and url:
+                        source_documents.append(f"""[{len(source_documents) + 1}] [{title}]({url})\n""")
         yield json.dumps({"docs": source_documents}, ensure_ascii=False) 
 
     return EventSourceResponse(knowledge_base_chat_iterator(query))