优化对话转工单功能，添加重试机制以提高稳定性，限制处理会话数量为前2000个，更新示例查询和文件路径，增强代码可读性和维护性。同时新增数据库客户端功能，支持批量处理会话数据并导出至Excel。

rag2_0/intent_recognition/IntentRecognition.py

@@ -14,6 +14,8 @@ import json
 from typing import List, Tuple, Dict, Any, Optional
 import re
 import jieba
+import time
+
 from .PromptTemplates import (classification_prompt, query_rewrite_prompt, 
                               extract_nouns_prompt, classification_info, 
                               slot_filling_prompt)
@@ -95,7 +97,9 @@ class IntentRecognizer:
         except Exception as e:
             raise RuntimeError(f"加载后缀关键词失败: {e}") from e
     
-    def _classify_intent(self, query: str) -> Classification:
+    def _classify_intent(self, query: str, conversation_context: str = "", 
+                      chat_history: List[Dict[str, str]] = None, 
+                      previous_slots: Dict[str, Any] = None) -> Classification:
         """
         对用户输入进行意图分类
         
@@ -109,7 +113,9 @@ class IntentRecognizer:
         classification_parser = PydanticOutputParser(pydantic_object=Classification)
         formatted_prompt = classification_prompt.format(user_input=query, 
                                                         classification_info=classification_info, 
-                                                        output_format=classification_parser.get_format_instructions())
+                                                        output_format=classification_parser.get_format_instructions(),
+                                                        conversation_context=conversation_context,
+                                                        chat_history=json.dumps(chat_history, ensure_ascii=False))
 
         # 调用LLM
         response = self._llm.invoke(formatted_prompt, False)
@@ -208,7 +214,7 @@ class IntentRecognizer:
             term_texts = ["名称:" + term.name + "|" + "同义词:" + ";".join(term.synonymous) for term in matched_terms]
             
             # 使用重排序模型
-            xinference_reranker = SiliconFlowReRankerModel()
+            xinference_reranker = XinferenceReRankerModel()
             rerank_results = xinference_reranker.rerank(query_key, term_texts, top_k=top_k)
             
             # 将matched_terms转换为列表以便按索引访问
@@ -220,7 +226,7 @@ class IntentRecognizer:
             return reranked_terms
             
         except Exception as e:
-            raise RuntimeError(f"SiliconFlowReRankerModel重排失败：{e}") from e
+            raise RuntimeError(f"_rerank_matched_terms重排失败：{e}") from e
 
     def _match_keywords(self, query: str, use_jieba: bool = False) -> Tuple[TermList, List[str]]:
         """
@@ -233,18 +239,23 @@ class IntentRecognizer:
         Returns:
             匹配到的关键词列表
         """
+        start_time = time.time()
         query_keys=[]
         # 步骤1: 使用LLM提取查询中的关键词
         try:
+            llm_start_time = time.time()
             extracted_terms = self._extract_keywords_with_llm(query, use_jieba)
             for term in extracted_terms:
                 query_keys.append(term.name)
+            llm_end_time = time.time()
+            llm_time = llm_end_time - llm_start_time
         except Exception as e:
             raise RuntimeError(f"LLM关键词提取失败: {e}") from e
         
         matched_terms = []  # 存储匹配到的Term对象       
         # 步骤2: 使用向量检索找到相似的专业名词
         try:
+            vector_start_time = time.time()
             # 对matched_terms中的每个关键字进行向量检索
             for current_key in query_keys:
                 vector_results = self._noun_retriever.query(current_key, top_k=5, use_intersection=False)
@@ -262,12 +273,20 @@ class IntentRecognizer:
                 if len(current_key_terms) > 0:
                     reranked_terms = self._rerank_matched_terms(current_key, current_key_terms)
                     matched_terms.extend(reranked_terms)
+            vector_end_time = time.time()
+            vector_time = vector_end_time - vector_start_time
         except Exception as e:
             raise RuntimeError(f"向量检索关键词时出错: {e}") from e
         
         # 提取所有Term对象的名称并排序
         # 将set类型的matched_terms转换为TermList类型
         term_list = TermList(terms=list(matched_terms))
+        end_time = time.time()
+        total_time = end_time - start_time
+
+        # 输出整合的时间日志
+        logging.info(f"关键词匹配耗时统计 - 总耗时: {total_time:.2f}秒, 问题关键词提取: {llm_time:.2f}秒, 向量检索+重排序: {vector_time:.2f}秒")
+        
         return term_list, query_keys
     
     def _rewrite_query(self, query: str, keywords: TermList, query_keys:List[str], chat_history: List[Dict[str, str]] = None, context: str = "") -> QueryRewrite:
@@ -282,6 +301,8 @@ class IntentRecognizer:
         Returns:
             改写结果
         """
+
+        rewrite_start_time = time.time()
         # 准备问题改写提示
         # terms_dict = [term.model_dump(exclude={"description"}) for term in keywords.terms] 
         terms_dict = [term.model_dump() for term in keywords.terms] 
@@ -295,7 +316,7 @@ class IntentRecognizer:
                                                            keywords=keywords_str,
                                                            chat_history=chat_history,
                                                            context=context)
-        
+
         # 调用LLM
         response = self._llm.invoke(formatted_prompt, False)
         
@@ -303,6 +324,9 @@ class IntentRecognizer:
         try:
             # 尝试直接解析JSON响应
             parsed_output = query_rewrite_parser.parse(response.content)
+            rewrite_end_time = time.time()
+            rewrite_time = rewrite_end_time - rewrite_start_time
+            logging.info(f"问题改写耗时统计 - 总耗时: {rewrite_time:.2f}秒")
             return parsed_output
         except Exception as e:
             raise RuntimeError(f"解析问题改写结果时出错: {e}") from e
@@ -360,7 +384,10 @@ class IntentRecognizer:
         #         suffix_terms.append(suffix_term)
    
         #     return Classification(vertical_classification="安装下载", sub_classification="查询"), TermList(terms=suffix_terms), QueryRewrite(rewrite=query), matched_suffixes
-
+        if chat_history is None:
+            chat_history = []
+        if previous_slots is None:
+            previous_slots = {}
         # 步骤1: 匹配关键词
         keywords_terms, query_keys = self._match_keywords(query, use_jieba)
 
@@ -397,7 +424,9 @@ class IntentRecognizer:
         # }
 
     
-    def _fill_slots(self, query: str, classification: Classification) -> Dict[str, Any]:
+    def _fill_slots(self, query: str, classification: Classification, conversation_context: str = "", 
+                      chat_history: List[Dict[str, str]] = None, 
+                      previous_slots: Dict[str, Any] = None,) -> Dict[str, Any]:
         """
         根据分类结果对问题进行槽位填充
         
@@ -415,7 +444,7 @@ class IntentRecognizer:
             raise RuntimeError("未找到匹配的槽位模型")
         
         # 使用LLM进行槽位填充
-        filled_slots = self._fill_slots_with_llm(query, classification, slot_model)
+        filled_slots = self._fill_slots_with_llm(query, classification, slot_model, conversation_context, chat_history, previous_slots)
         
         # 检查必填槽位是否都已填充
         is_complete, missing_slots = filled_slots.check_required_slots()
@@ -467,7 +496,12 @@ class IntentRecognizer:
             
         return None
 
-    def _fill_slots_with_llm(self, query: str, classification: Classification, slot_model_class: type) -> Any:
+    def _fill_slots_with_llm(self, query: str, 
+                             classification: Classification, 
+                             slot_model_class: type, 
+                             conversation_context: str = "", 
+                            chat_history: List[Dict[str, str]] = None, 
+                            previous_slots: Dict[str, Any] = None) -> Any:
         """
         使用LLM进行槽位填充
         
@@ -486,7 +520,10 @@ class IntentRecognizer:
             query=query,
             vertical_classification=classification.vertical_classification,
             sub_classification=classification.sub_classification,
-            output_format=slot_parser.get_format_instructions()
+            output_format=slot_parser.get_format_instructions(),
+            conversation_context=conversation_context,
+            chat_history=json.dumps(chat_history,ensure_ascii=False),
+            previous_slots=json.dumps(previous_slots,ensure_ascii=False),
         )
         
         # 调用LLM
@@ -537,9 +574,14 @@ class IntentRecognizer:
             output_format=parser.get_format_instructions(),
             classification_info=classification_info
         )
+        
         # 调用LLM
+        llm_start_time = time.time()
         response = self._llm.invoke(formatted_prompt + output_example, False)
+        llm_end_time = time.time()
+        llm_time = llm_end_time - llm_start_time
 
+        
         try:
             # 解析LLM响应为JSON
             result_json = parser.parse(response.content)
@@ -552,8 +594,19 @@ class IntentRecognizer:
             if expected_slot_model is None:
                 # 添加容错处理，应对LLM返回错误分类信息,一级分类跟二级分类错乱
                 # 重新分类
-                classification = self._classify_intent(user_input)
-                fill_slots = self._fill_slots(user_input, classification)
+                classify_start_time = time.time()
+                classification = self._classify_intent(user_input, conversation_context, chat_history, previous_slots)
+                classify_end_time = time.time()
+                classify_time = classify_end_time - classify_start_time
+                # logging.info(f"重新分类耗时: {classify_time:.2f}秒")
+                
+                fill_start_time = time.time()
+                fill_slots = self._fill_slots(user_input, classification, conversation_context, chat_history, previous_slots)
+                fill_end_time = time.time()
+                fill_time = fill_end_time - fill_start_time
+                all_time=fill_end_time-llm_start_time
+                logging.info(f"总耗时：{all_time:.2f}秒,首次槽位+分类:{llm_time:.2f}秒, 重新分类耗时: {classify_time:.2f}秒, 重新槽位填充耗时: {fill_time:.2f}秒")
+                
                 result = {
                     "classification": classification.model_dump(),
                     "slot_filling": fill_slots
@@ -562,13 +615,21 @@ class IntentRecognizer:
                 return result
             elif expected_slot_model.__name__ !=  type(slot_filling).__name__:
                 # 添加容错处理，应对LLM槽位与分类不匹配。重新填充槽位
+                fill_start_time = time.time()
                 slot_filling = self._fill_slots(user_input, classification)
+                fill_end_time = time.time()
+                fill_time = fill_end_time - fill_start_time
+                all_time=fill_end_time-llm_start_time
+                logging.info(f"总耗时：{all_time:.2f}秒,首次槽位+分类:{llm_time:.2f}秒, 重新槽位填充耗时: {fill_time:.2f}秒")
+
                 result = {
                     "classification": classification.model_dump(),
                     "slot_filling": slot_filling
                 }
                 logging.warning(f"重新填充槽点")
                 return result
+            
+            logging.info(f"意图识别+槽位LLM调用耗时: {llm_time:.2f}秒")
 
             # 构建最终结果
             result = {