feat(knowledge): 优化知识库返回结构和集成流程

- 重构QAManager.get_knowledge()返回结构化字典包含知识项和摘要
- 在bot.py中添加initialize_lpmm_knowledge()初始化调用
- 更新Prompt类知识库集成逻辑，改进错误处理
- 优化lpmm知识工具显示格式，增强用户体验
- 调整异步任务基础超时时间从10秒增至20秒

bot.py

@@ -21,6 +21,7 @@ initialize_logging()
 from src.main import MainSystem  # noqa
 from src import BaseMain  # noqa
 from src.manager.async_task_manager import async_task_manager  # noqa
+from src.chat.knowledge.knowledge_lib import initialize_lpmm_knowledge # noqa
 from src.config.config import global_config  # noqa
 from src.common.database.database import initialize_sql_database  # noqa
 from src.common.database.sqlalchemy_models import initialize_database as init_db  # noqa
@@ -228,6 +229,7 @@ if __name__ == "__main__":
         try:
             # 执行初始化和任务调度
             loop.run_until_complete(main_system.initialize())
+            initialize_lpmm_knowledge()
             # Schedule tasks returns a future that runs forever.
             # We can run console_input_loop concurrently.
             main_tasks = loop.create_task(main_system.schedule_tasks())

src/chat/knowledge/qa_manager.py

@@ -1,5 +1,5 @@
 import time
-from typing import Tuple, List, Dict, Optional
+from typing import Tuple, List, Dict, Optional, Any
 
 from .global_logger import logger
 from .embedding_store import EmbeddingManager
@@ -98,30 +98,46 @@ class QAManager:
 
         return result, ppr_node_weights
 
-    async def get_knowledge(self, question: str) -> Optional[str]:
+    async def get_knowledge(self, question: str) -> Optional[Dict[str, Any]]:
-        """获取知识"""
+        """
-        # 处理查询
+        获取知识，返回结构化字典
-        processed_result = await self.process_query(question)
-        if processed_result is not None:
-            query_res = processed_result[0]
-            # 检查查询结果是否为空
-            if not query_res:
-                logger.debug("知识库查询结果为空，可能是知识库中没有相关内容")
-                return None
         
-            knowledge = [
+        Args:
-                (
+            question: 用户提出的问题
-                    self.embed_manager.paragraphs_embedding_store.store[res[0]].str,
+
-                    res[1],
+        Returns:
-                )
+            一个包含 'knowledge_items' 和 'summary' 的字典，或者在没有结果时返回 None
-                for res in query_res
+        """
-            ]
+        processed_result = await self.process_query(question)
-            found_knowledge = "\n".join(
+        if not processed_result or not processed_result[0]:
-                [f"第{i + 1}条知识：{k[0]}\n 该条知识对于问题的相关性：{k[1]}" for i, k in enumerate(knowledge)]
+            logger.debug("知识库查询结果为空。")
-            )
-            if len(found_knowledge) > MAX_KNOWLEDGE_LENGTH:
-                found_knowledge = found_knowledge[:MAX_KNOWLEDGE_LENGTH] + "\n"
-            return found_knowledge
-        else:
-            logger.debug("LPMM知识库并未初始化，可能是从未导入过知识...")
             return None
+
+        query_res = processed_result[0]
+        
+        knowledge_items = []
+        for res_hash, relevance, *_ in query_res:
+            if store_item := self.embed_manager.paragraphs_embedding_store.store.get(res_hash):
+                knowledge_items.append({
+                    "content": store_item.str,
+                    "source": "内部知识库",
+                    "relevance": f"{relevance:.4f}"
+                })
+
+        if not knowledge_items:
+            return None
+            
+        # 使用LLM生成总结
+        knowledge_text_for_summary = "\n\n".join([item['content'] for item in knowledge_items[:5]]) # 最多总结前5条
+        summary_prompt = f"根据以下信息，为问题 '{question}' 生成一个简洁的、不超过50字的摘要：\n\n{knowledge_text_for_summary}"
+        
+        try:
+            summary, (_, _, _) = await self.qa_model.generate_response_async(summary_prompt)
+        except Exception as e:
+            logger.error(f"生成知识摘要失败: {e}")
+            summary = "无法生成摘要。"
+
+        return {
+            "knowledge_items": knowledge_items,
+            "summary": summary.strip() if summary else "没有可用的摘要。"
+        }

src/chat/utils/prompt.py

@@ -369,7 +369,7 @@ class Prompt:
                 task_names.append("cross_context")
             
             # 性能优化
-            base_timeout = 10.0
+            base_timeout = 20.0
             task_timeout = 2.0
             timeout_seconds = min(
                 max(base_timeout, len(tasks) * task_timeout),
@@ -676,22 +676,21 @@ class Prompt:
             return {"knowledge_prompt": ""}
         
         try:
-            from src.chat.knowledge.knowledge_lib import QAManager
+            from src.chat.knowledge.knowledge_lib import qa_manager
             
             # 获取问题文本（当前消息）
             question = self.parameters.target or ""
             if not question:
                 return {"knowledge_prompt": ""}
             
-            # 创建QA管理器
+            # 检查QA管理器是否已成功初始化
-            qa_manager = QAManager()
+            if not qa_manager:
+                logger.warning("QA管理器未初始化 (可能lpmm_knowledge被禁用)，跳过知识库搜索。")
+                return {"knowledge_prompt": ""}
             
             # 搜索相关知识
             knowledge_results = await qa_manager.get_knowledge(
-                question=question,
+                question=question
-                chat_id=self.parameters.chat_id,
-                max_results=5,
-                min_similarity=0.5
             )
             
             # 构建知识块
@@ -704,13 +703,10 @@ class Prompt:
                     relevance = item.get("relevance", 0.0)
                     
                     if content:
-                        if source:
+                        knowledge_parts.append(f"- [相关度: {relevance}] {content}")
-                            knowledge_parts.append(f"- [{relevance:.2f}] {content} (来源: {source})")
-                        else:
-                            knowledge_parts.append(f"- [{relevance:.2f}] {content}")
                 
-                if knowledge_results.get("summary"):
+                if summary := knowledge_results.get("summary"):
-                    knowledge_parts.append(f"\n知识总结: {knowledge_results['summary']}")
+                    knowledge_parts.append(f"\n知识总结: {summary}")
                 
                 knowledge_prompt = "\n".join(knowledge_parts)
             else:

src/plugins/built_in/knowledge/lpmm_get_knowledge.py

@@ -43,10 +43,16 @@ class SearchKnowledgeFromLPMMTool(BaseTool):
 
             logger.debug(f"知识库查询结果: {knowledge_info}")
 
-            if knowledge_info:
+            if knowledge_info and knowledge_info.get("knowledge_items"):
-                content = f"你知道这些知识: {knowledge_info}"
+                knowledge_parts = []
+                for i, item in enumerate(knowledge_info["knowledge_items"]):
+                    knowledge_parts.append(f"- {item.get('content', 'N/A')}")
+                
+                knowledge_text = "\n".join(knowledge_parts)
+                summary = knowledge_info.get('summary', '无总结')
+                content = f"关于 '{query}', 你知道以下信息：\n{knowledge_text}\n\n总结: {summary}"
             else:
-                content = f"你不太了解有关{query}的知识"
+                content = f"关于 '{query}'，你的知识库里好像没有相关的信息呢"
             return {"type": "lpmm_knowledge", "id": query, "content": content}
         except Exception as e:
             # 捕获异常并记录错误