更新对话到工单的处理逻辑，添加API密钥参数，优化响应内容清理，修复多个解析函数中的API调用，调整并发处理的最大线程数。

rag2_0/demo/dialogue_to_workorder.py

@@ -135,15 +135,23 @@ class DialogueToWorkorder:
         self.product_line_parser = PydanticOutputParser(pydantic_object=ProductLine)
         
         # 初始化LLM模型
+        # self.llm_params = llm_params or {
+        #     "temperature": 0.2,
+        #     "top_p":0.95,
+        #     "model": "deepseek-ai/DeepSeek-R1",
+        #     "api_key": os.getenv("OPENAI_API_KEY"),
+        #     "base_url": os.getenv("OPENAI_API_BASE"),
+        #     "timeout": httpx.Timeout(600.0)
+        # }
+        self.api_key = "25t%Syu6I9yxX2IuTN"
         self.llm_params = llm_params or {
             "temperature": 0.2,
             "top_p":0.95,
-            "model": "deepseek-ai/DeepSeek-R1",
-            "api_key": os.getenv("OPENAI_API_KEY"),
-            "base_url": os.getenv("OPENAI_API_BASE"),
+            "model": "deepseek-r1",
+            "api_key": "25t%Syu6I9yxX2IuTN",
+            "base_url": "http://10.1.0.154:8000/v1",
             "timeout": httpx.Timeout(600.0)
         }
-        
         self.llm = self._get_llm_instance()
     
     def _get_llm_instance(self):
@@ -266,7 +274,7 @@ class DialogueToWorkorder:
 2、根据用户提出的问题，分析坐席提供的解决方法(比如：1、引导用户xxxx。2、告诉用户xxxxx)。以坐席的角度直接总结完整的解决方案或应对措施(不要出现"坐席"、"我"等字样)
 3、提炼访客独立的核心问题(以访客的角度总结核心问题)，核心问题衍生、细化后的请求合并到对应的核心问题中。不要单独列出衍生、细化后的请求。
 
-4、使用json格式输出(多个用户问题采用标准json数组格式输出)：
+4、使用json格式输出(多个用户问题采用标准json格式输出)：
 {output_format}
 
 输出示例：
@@ -283,15 +291,15 @@ class DialogueToWorkorder:
         output_format = self.user_question_and_solution_parser.get_format_instructions()
         llm_prompt = prompt.format(output_format=output_format, dialogue_str=dialogue_str)
         
-        response = self.llm.invoke(user_prompt=llm_prompt, need_retry=False)
-        
+        response = self.llm.invoke(user_prompt=llm_prompt, need_retry=False, api_key=self.api_key)
+        clean_output = re.sub(r'<think>.*?</think>', '', response.content, flags=re.DOTALL)
         try:
-            if response.content.count('user_question') == 1:
-                user_question_and_solution = self.user_question_and_solution_parser.parse(response.content)
+            if clean_output.count('user_question') == 1:
+                user_question_and_solution = self.user_question_and_solution_parser.parse(clean_output)
                 return [user_question_and_solution]
             else:
                 array_pattern = r'\[\s*(\{[\s\S]*?\}(?:\s*,\s*\{[\s\S]*?\})*)\s*\]'
-                array_match = re.search(array_pattern, response.content)
+                array_match = re.search(array_pattern, clean_output)
                 if array_match:
                     # 找到了JSON数组
                     json_array_str = '[' + array_match.group(1) + ']'
@@ -313,8 +321,9 @@ class DialogueToWorkorder:
         except Exception as e:
             output_format = self.user_question_and_solution_list_parser.get_format_instructions()
             llm_prompt = prompt.format(output_format=output_format, dialogue_str=dialogue_str)
-            response = self.llm.invoke(user_prompt=llm_prompt, need_retry=False)
-            user_question_and_solution_temp = self.user_question_and_solution_list_parser.parse(response.content)
+            response = self.llm.invoke(user_prompt=llm_prompt, need_retry=False, api_key=self.api_key)
+            clean_output = re.sub(r'<think>.*?</think>', '', response.content, flags=re.DOTALL)
+            user_question_and_solution_temp = self.user_question_and_solution_list_parser.parse(clean_output)
             return user_question_and_solution_temp.user_question_list
 
         return [user_question_and_solution]
@@ -345,8 +354,9 @@ class DialogueToWorkorder:
 {dialogue_str}
         """
         
-        response = self.llm.invoke(user_prompt=prompt, need_retry=False)
-        product_name_and_module_name = self.product_name_and_module_name_parser.parse(response.content)
+        response = self.llm.invoke(user_prompt=prompt, need_retry=False, api_key=self.api_key)
+        clean_output = re.sub(r'<think>.*?</think>', '', response.content, flags=re.DOTALL)
+        product_name_and_module_name = self.product_name_and_module_name_parser.parse(clean_output)
         
         return product_name_and_module_name.product_name, product_name_and_module_name.module_name
     
@@ -374,8 +384,9 @@ class DialogueToWorkorder:
 {dialogue_str}
         """
         
-        response = self.llm.invoke(user_prompt=prompt, need_retry=False)
-        product_line = self.product_line_parser.parse(response.content)
+        response = self.llm.invoke(user_prompt=prompt, need_retry=False, api_key=self.api_key)
+        clean_output = re.sub(r'<think>.*?</think>', '', response.content, flags=re.DOTALL)
+        product_line = self.product_line_parser.parse(clean_output)
         
         return product_line.product_line
     
@@ -410,8 +421,9 @@ class DialogueToWorkorder:
 {dialogue_str}
         """
         
-        response = self.llm.invoke(user_prompt=prompt, need_retry=False)
-        question_type = self.question_type_parser.parse(response.content)
+        response = self.llm.invoke(user_prompt=prompt, need_retry=False, api_key=self.api_key)
+        clean_output = re.sub(r'<think>.*?</think>', '', response.content, flags=re.DOTALL)
+        question_type = self.question_type_parser.parse(clean_output)
         
         return question_type.question_type
     
@@ -446,8 +458,9 @@ class DialogueToWorkorder:
 
         """
         
-        response = self.llm.invoke(user_prompt=prompt, need_retry=False)
-        is_complaint = self.is_complaint_parser.parse(response.content)
+        response = self.llm.invoke(user_prompt=prompt, need_retry=False, api_key=self.api_key)
+        clean_output = re.sub(r'<think>.*?</think>', '', response.content, flags=re.DOTALL)
+        is_complaint = self.is_complaint_parser.parse(clean_output)
         
         return (is_complaint.is_dissatisfaction, 
                 is_complaint.dissatisfaction_level, 
@@ -458,6 +471,7 @@ class DialogueToWorkorder:
         """处理单个会话的函数，用于多线程并发"""
         # if conversation_id!="b157aa91-3acb-11f0-a191-4fb224ef4b40":
         #     return []
+        try:
             # 获取工单基本信息
             base_workorder_dict = self.get_workorder_dict(conversation_rows)
             # 分析用户问题和解决方案
@@ -533,6 +547,9 @@ class DialogueToWorkorder:
                 workorder_list.append(workorder_dict)
             
             return workorder_list
+        except Exception as e:
+            logger.error(f"处理会话ID: {conversation_id} 时发生错误: {e}")
+            return []
     
     def analyze_conversation_data(self, conversation_excel_path, product_detail_excel_path, max_workers=10, start_date=None, end_date=None):
         """分析会话数据主流程，使用多线程并发处理"""
@@ -588,6 +605,7 @@ class DialogueToWorkorder:
                 conversation_id = future_to_conversation[future]
                 try:
                     result_workorders = future.result()
+                    if result_workorders:
                         # 将每个会话的所有工单添加到总列表中
                         workorder_dict_list.extend(result_workorders)
                         logger.info(f"完成处理会话ID: {conversation_id}，生成工单数量: {len(result_workorders)}")
@@ -678,7 +696,7 @@ def parse_arguments():
                         help='会话内容Excel文件路径')
     parser.add_argument('--product_detail_file', type=str, required=False,
                         help='产品详情Excel文件路径')
-    parser.add_argument('--max_workers', type=int, default=16,
+    parser.add_argument('--max_workers', type=int, default=6,
                         help='并发处理线程数，默认为16')
     parser.add_argument('--start_date', type=str, required=False,default="2025-06-10 16:08:00",
                         help='开始日期，格式为YYYY-MM-DD')