更新多个模块，优化API密钥管理，增强意图识别和Excel数据验证功能，添加置信度评分，改进日志记录和错误处理机制，支持调试模式，提升代码可读性和灵活性。

api_key.txt

@@ -1,11 +1,7 @@
 sk-kvgfuqeqvpmfsccykyoohheshclcrtvjlnewratvrjpkpbkc
-sk-mxbapcczwjsyrictwgigxgcvdgptyfrlynrewqioegqwrggv
-sk-dujjzxrknevesbagqgqmuffxsosjoueviubnmodoormlmlzt
-sk-lsukfggzghmdhtfhqcbmlfqabbtapwpuxnvtwshqqqlaesie
 sk-aulumxzhvaladchcwgmsxidtdsvzytbpvzqgfuvcxlwbwcgl
 sk-otxxemniwhxkdvroszmmkitswwuykosnqoldrkzdoflqpgvw
 sk-zlruqobfdbjebyyvkmehakpcvfgnlfbdlbfrepusazzckbnv
-sk-zryimztrlkgvcaiolarhvbcewmhwruhqfcndbylonzlqvdox
 sk-rczjqufgdisqplkrmvhaxmdgcboluvxympvzljlreuqeeviq
 sk-xfnvcksdgwufsktvmhpqrwpgovsxxtaeehtxnaqjtxmubqzl
 sk-wwonvjnowbcxmoyoluynnkjwerghspzdulyidskunkordaft
@@ -16,10 +12,8 @@ sk-lvtdgodiaurqyiwdxtdrgxifguychhccqlqkhqctscvqbfgi
 sk-aedlbtlmqcttxwnvlfmxzaysamocamqxjceoyqjfgpcowybw
 sk-fahdvndjblyjlizamvwcrxnilsgmbgbvwssxgquhkezgpqne
 sk-tzludgttzxvpvwayazdbppbauvathdtccafjrhojpemucgyi
-sk-hrbroidbfusidwnsmxenuzljxgdzzxiimlezygxplavnxjik
 sk-ylgoiqxmtxeojdnonthxtweungyzldaqarvjxlqyztlvyrff
 sk-asuqbqwdhjcqnvtjlwufyrkrwkobnrbmukzarvcctsgjipdp
-sk-dpgpymiydutoexgvkajwgahagnfmcqzafwulccudnzvleifz
 sk-nbksjgcngsayoumnsdbkcpnqivnvxjenwpzuazzrkhnsgeoo
 sk-jgybgyayxlwoxeijgrjcneqlyusleohgbliuwpsuhocrjsmk
 sk-wzjsmwxcbbpcrqivqfzjwufqqjtlwejtncnvbpeicznkwiuh
@@ -28,7 +22,6 @@ sk-fcsfmyivfuojsqsditvobfqprdpeunukycpcfnoxkraqevpx
 sk-szyjgyxrcvyxpvzfwgmbxnflxngxvcplitcctsdvvrqjgftk
 sk-jzbodthsnvjwbyrnynsxrudtqfnbdbrcxebjwjgajocnzqse
 sk-fxepossfzpmccibfwqpkluorzqlbtcaplepeugtfzfsctcbl
-sk-ympnflocrkxjrbubsxqdjqwicuyavvvysctlpfhunkcrzxjx
 sk-flhqvziknntednkcgjaxlyzzsrfzjhrzrmteqonajpbiinni
 sk-xfregpbbquqbxpiobjzanydsjivrjrnbokzxcqtnhxhyghhe
 sk-jrdzerhmvrtvzawkksowbgkggkubwfquplmrxbdhespqgtis
@@ -113,3 +106,13 @@ sk-yyqtojpqfrtkyvmhbjvnzhuujzgvitqpuxkytgxspwqscptz
 sk-vcwenoaiegiwqhdvalxrvttwqmcrudttpfqhlvtdocsxvvob
 sk-npnyztbrdaqtbnldlvfelkfbfozykuwykkkfsrsisytusymj
 sk-rjaadeqsgoskclkuqkkyjxdknngrdsjwlgucqutiskroprzk
+sk-uollmeyatyiwfzszvxkpyndmzfrbqjpyixewmrastbmaqbhy
+sk-xdlsjytiwilvodadkjxvwdgulhhdytkqvfpyrcnllclgzqkb
+sk-ffkltifkylutornjhwmnmfjsqsywrjibvujhjtjctzgnkvlp
+sk-vmwocqqjqxnsvzmeyvqskahjaclifpmsbhywvnrvwygkfyuj
+sk-gzwkmzxeeunaywrdrgirdatqhdtqdgvzqpesvprwbbjhcchn
+sk-duchutcxmygrnkhzmmlykvtzwaylqtdxfbbuhvfvzuapazii
+sk-nlddwexmjxqtgdvahwvlotnomrzcgskxeakxkxauicknzfkp
+sk-lopwluipwvilwpwztvaxfebueeyilefwgncgpeprqvwazxom
+sk-rgwrklpvhhrluokkbgavzukuhhpfhqzmozpjzoezfhkxyorc
+sk-cdrpglnfmyeeqyhtvxvkpcpwscsbfouwkagjpphuksfzeipy

rag2_0/demo/dialogue_to_workorder.py

@@ -8,13 +8,17 @@ import concurrent.futures
 from functools import wraps
 from pydantic import BaseModel, Field
 from langchain.output_parsers import PydanticOutputParser
-from rag2_0.tool.ModelTool import OpenAiLLM
+import sys
 from dotenv import load_dotenv
 import httpx
 import traceback
 import re
 import logging
 
+# 将项目根目录添加到Python路径
+sys.path.append(os.getcwd())
+from rag2_0.tool.ModelTool import OpenAiLLM
+
 load_dotenv()
 
 # 配置日志
@@ -31,7 +35,7 @@ logger = logging.getLogger("dialogue_to_workorder")
 # ================ 模型定义 ================
 class UserQuestionAndSolution(BaseModel):
     user_question: str = Field(description="用户的核心问题")
-    solution: str = Field(description="坐席提供的解决方案")
+    solution: str = Field(description="坐席提供的解决方案,解决方案如果存在多个步骤，使用中文分号隔开")
 
 class UserQuestionAndSolutionList(BaseModel):
     user_question_list: list[UserQuestionAndSolution] = Field(description="客户问题列表")
@@ -236,7 +240,7 @@ class DialogueToWorkorder:
 ```json
 {{
   "user_question": "技改软件打开报错",
-  "solution": "1、告知报错原因 2、通过远程辅助解决"
+  "solution": "1、告知报错原因；2、通过远程辅助解决"
 }} 
 ```
 =======对话记录如下所示=======
@@ -426,14 +430,15 @@ class DialogueToWorkorder:
         # 分析用户问题和解决方案
         user_question_list = self.get_user_question_and_solution(conversation_rows)
 
-        # 获取第一个问题和解决方案，用于后续分析
+        user_question_str=""
-        if user_question_list and len(user_question_list) > 0:
+        for user_question in user_question_list:
-            first_question = user_question_list[0]
+            user_question_str = user_question_str + user_question.user_question.strip() + "\n"
-            user_question_str = first_question.user_question
+        user_question_str = user_question_str.strip()
-            solution_str = first_question.solution
+
-        else:
+        solution_str=""
-            user_question_str = ""
+        for user_question in user_question_list:
-            solution_str = ""
+            solution_str = solution_str + user_question.solution.strip() + "\n"
+        solution_str = solution_str.strip()
 
         # 分析是否抱怨、是否投诉、抱怨级别
         is_dissatisfaction, dissatisfaction_level, dissatisfaction_reasoning, is_complaint = (
@@ -455,15 +460,9 @@ class DialogueToWorkorder:
         # 创建工单列表
         workorder_list = []
         
-        for user_question in user_question_list:
-            user_question_str = user_question.user_question
-            solution_str = user_question.solution
 
-            # 创建新的工单字典，复制基本信息
+        # 更新工单字典
-            workorder_dict = base_workorder_dict.copy()
+        base_workorder_dict.update({
-
-            # 更新工单字典
-            workorder_dict.update({
                 "产品线": product_line,
                 "产品名称": product_name,
                 "模块名称": module_name,
@@ -474,9 +473,29 @@ class DialogueToWorkorder:
                 "是否投诉": "是" if is_complaint else '否',
                 "解决方案": (solution_str + '\n存在抱怨:' + dissatisfaction_reasoning) if is_dissatisfaction else solution_str
             })
+        workorder_list.append(base_workorder_dict)
+        # for user_question in user_question_list:
+        #     user_question_str = user_question.user_question
+        #     solution_str = user_question.solution
             
-            # 将工单添加到列表中
+        #     # 创建新的工单字典，复制基本信息
-            workorder_list.append(workorder_dict)
+        #     workorder_dict = base_workorder_dict.copy()
+
+        #     # 更新工单字典
+        #     workorder_dict.update({
+        #         "产品线": product_line,
+        #         "产品名称": product_name,
+        #         "模块名称": module_name,
+        #         "客户问题": user_question_str,
+        #         "问题类型": problem_type,
+        #         "是否抱怨": "是" if is_dissatisfaction else '否',
+        #         "抱怨级别": dissatisfaction_level if is_dissatisfaction else '',
+        #         "是否投诉": "是" if is_complaint else '否',
+        #         "解决方案": (solution_str + '\n存在抱怨:' + dissatisfaction_reasoning) if is_dissatisfaction else solution_str
+        #     })
+            
+        #     # 将工单添加到列表中
+        #     workorder_list.append(workorder_dict)
         
         return workorder_list
     

rag2_0/demo/intent_recognition_example.py

@@ -8,8 +8,6 @@ Description: 意图识别和问题改写示例
 
 import os
 from dotenv import load_dotenv
-from regex import F
-from rag2_0.intent_recognition import IntentRecognizer
 import pandas as pd
 import logging
 import json
@@ -19,6 +17,8 @@ import time
 import sys
 import argparse
 from typing import List, Dict
+sys.path.append(os.getcwd())
+from rag2_0.intent_recognition import IntentRecognizer
 # 加载环境变量
 load_dotenv()
 
@@ -176,7 +176,7 @@ def save_results_to_excel(results, output_file, is_final=False):
     logging.info(f"已保存{len(valid_results)}条结果至: {temp_output_file}")
 
 # 示例查询
-examples_query = """.BDD3是哪款软件编制的"""
+examples_query = """D3软件结算工程怎么解锁清单"""
 conversation_context=""
 chat_history=[
         {

rag2_0/demo/validate_excel_data_batch.py

@@ -6,6 +6,7 @@ Description: 使用LLM批量验证Excel数据中的问题分类、问题拆解
 """
 
 import os
+import sys
 import pandas as pd
 import json
 import argparse
@@ -14,13 +15,25 @@ import concurrent.futures
 from tqdm import tqdm
 from dotenv import load_dotenv
 from langchain_openai import ChatOpenAI
-from rag2_0.intent_recognition.PromptTemplates import classification
+from pydantic import BaseModel, Field
+from langchain.output_parsers import PydanticOutputParser
+
+sys.path.append(os.getcwd())
+from rag2_0.intent_recognition.PromptTemplates import classification_info
+from rag2_0.intent_recognition.DataModels import *
 from rag2_0.tool.ModelTool import OpenAiLLM
 
+
+# 定义验证结果的Pydantic模型
+class ValidationResult(BaseModel):
+    is_correct: bool = Field(description="验证是否通过")
+    confidence_score: float = Field(description="置信度得分")
+    reason: str = Field(default="", description="得出结论的原因")
+
 class ExcelDataValidator:
     """Excel数据验证类，用于批量验证Excel数据中的问题分类、问题拆解、检索关键词和问题改写"""
     
-    def __init__(self, input_file=None, output_file=None, workers=4, batch_size=10):
+    def __init__(self, input_file=None, output_file=None, workers=4, batch_size=10, debug=False):
         """
         初始化验证器
         
@@ -29,6 +42,7 @@ class ExcelDataValidator:
             output_file: 输出结果Excel文件路径
             workers: 并行工作线程数
             batch_size: 每批处理的行数
+            debug: 是否启用调试模式（串行处理）
         """
         # 加载环境变量
         load_dotenv()
@@ -37,6 +51,7 @@ class ExcelDataValidator:
         self.output_file = output_file
         self.workers = workers
         self.batch_size = batch_size
+        self.debug = debug
         self.df = None
         
         # 设置日志
@@ -71,7 +86,7 @@ class ExcelDataValidator:
             
         try:
             df = pd.read_excel(file_path)
-            required_columns = ["提问", "问题拆解", "一级分类", "二级分类", "问题改写", "检索的关键词"]
+            required_columns = ["问题", "问题分类", "问题改写", "槽点信息", "检索的内容"]
             for col in required_columns:
                 if col not in df.columns:
                     logging.error(f"缺少必要的列: {col}")
@@ -94,109 +109,167 @@ class ExcelDataValidator:
             sub_class: 二级分类
             
         Returns:
-            (bool, str): 是否正确，错误原因(如果有)
+            (bool, str, float): 是否正确，错误原因(如果有)，置信度
         """
+        parser = self.create_validation_parser()
+        format_instructions = parser.get_format_instructions()
+        
         prompt = f"""
         背景：用户正在使用电力造价软件，提出的问题可能涉及电力造价软件的使用，也可能涉及电力造价专业知识。我对用户问题进行了分类，请评估以下问题分类是否正确。
 
 我目前总共有以下分类：
-{classification}
+{classification_info}
 
 问题的分类情况如下：
 原始问题: {query}
 一级分类: {vertical_class}
 二级分类: {sub_class}
 
-请从专业角度分析这个分类是否准确。只需返回"正确"或"错误:原因"，不需要其他解释。"""
+请从专业角度分析这个分类是否准确，并以JSON格式返回结果。请提供一个0到1之间的置信度得分，表示你对判断的确信程度。
+
+{format_instructions}
+"""
 
         try:
             response = llm.invoke(prompt)
-            result = response.content.strip()
+            result = parser.parse(response.content)
-            
+            return result.is_correct, result.reason, result.confidence_score
-            if result.startswith("正确"):
-                return True, ""
-            else:
-                error_reason = result.replace("错误:", "").strip() if "错误:" in result else result
-                return False, error_reason
         except Exception as e:
             logging.warning(f"验证问题分类时出错: {e}")
-            return False, f"验证过程出错: {str(e)}"
+            return False, f"验证过程出错: {str(e)}", 0.0
     
-    def validate_query_keys(self, llm, query, query_keys):
+    def _get_slot_model(self, classification: Classification) -> Optional[type]:
         """
-        验证问题拆解是否正确
+        根据分类结果获取对应的槽位模型类，用于统一提示词处理
+        
+        Args:
+            classification: 意图分类结果
+            
+        Returns:
+            对应的槽位模型类
+        """
+        # 软件问题
+        if classification.vertical_classification == "软件问题":
+            if classification.sub_classification == "软件功能":
+                return SoftwareFunctionSlots
+            elif classification.sub_classification == "故障排查":
+                return SoftwareTroubleShootingSlots
+        
+        # 业务问题
+        elif classification.vertical_classification == "业务问题":
+            if classification.sub_classification == "专业咨询":
+                return ProfessionalConsultingSlots
+            elif classification.sub_classification == "数据问题":
+                return DataProblemSlots
+        
+        # 安装下载注册
+        elif classification.vertical_classification == "安装下载注册":
+            if classification.sub_classification == "后缀名咨询":
+                return FileExtensionConsultingSlots
+            elif classification.sub_classification == "软件锁类":
+                return SoftwareLockSlots
+            elif classification.sub_classification == "安装下载类":
+                return InstallationDownloadSlots
+            elif classification.sub_classification == "问题排查类":
+                return ProblemDiagnosisSlots
+        
+        # 其他
+        elif classification.vertical_classification == "其他":
+            return OtherSlots
+            
+        return None
+
+    def validate_slot(self, llm, rewrite, slot_info, vertical_class, sub_class):
+        """
+        验证槽位填充是否正确
         
         Args:
             llm: LLM模型
-            query: 原始问题
+            rewrite: 问题改写
-            query_keys: 问题拆解
+            slot_info: 槽位信息(JSON字符串)
             
         Returns:
-            (bool, str): 是否正确，错误原因(如果有)
+            (bool, str, float): 是否正确，错误原因(如果有)，置信度
         """
-        prompt = f"""
+        # 解析槽位信息JSON
-        背景：用户正在使用电力造价软件，提出的问题可能涉及电力造价软件的使用帮助，也可能涉及电力造价专业知识。我对用户问题进行了拆解，请评估以下问题拆解是否正确。
-
-原始问题: {query}
-问题拆解: {query_keys}
-
-问题拆解应该准确提取原始问题中的关键词和信息。请分析这个拆解是否准确。只需返回"正确"或"错误:原因"，不需要其他解释。"""
-
         try:
-            response = llm.invoke(prompt)
+            if isinstance(slot_info, str) and slot_info.strip():
-            result = response.content.strip()
+                slots = json.loads(slot_info)
-            
-            if result.startswith("正确"):
-                return True, ""
             else:
-                error_reason = result.replace("错误:", "").strip() if "错误:" in result else result
+                slots = slot_info
-                return False, error_reason
-        except Exception as e:
-            logging.warning(f"验证问题拆解时出错: {e}")
-            return False, f"验证过程出错: {str(e)}"
-    
-    def validate_keywords(self, llm, query, query_keys, keywords_str):
-        """
-        验证检索关键词是否准确
-        
-        Args:
-            llm: LLM模型
-            query: 原始问题
-            query_keys: 问题拆解
-            keywords_str: 检索关键词(JSON字符串)
-            
-        Returns:
-            (bool, str): 是否正确，错误原因(如果有)
-        """
-        # 解析关键词JSON
-        try:
-            if isinstance(keywords_str, str) and keywords_str.strip():
-                keywords = json.loads(keywords_str)
-            else:
-                keywords = []
         except:
-            keywords = keywords_str
+            slots = slot_info
         
+        parser = self.create_validation_parser()
+        format_instructions = parser.get_format_instructions()
+        slot_info_prompt = self._get_slot_model(Classification(vertical_classification=vertical_class, sub_classification=sub_class)).model_json_schema()
+        slot_info_prompt = json.dumps(slot_info_prompt, ensure_ascii=False)
         prompt = f"""
-    背景：用户正在使用电力造价软件，提出的问题可能涉及电力造价软件的使用帮助，也可能涉及电力造价专业知识。通过问题检索出了一些关键词，请评估这些关键词是否准确，是否与问题相关
+        背景：用户正在使用电力造价软件，提出的问题可能涉及电力造价软件的使用帮助，也可能涉及电力造价专业知识。我从用户问题中提取了槽位信息，请评估这些槽位信息是否准确、完整。
-原始问题: {query}
-问题拆解: {query_keys}
-检索关键词: {keywords}
 
-检索关键词应该准确反映问题中需要检索的关键概念和术语。请分析这些关键词是否准确、完整。只需返回"正确"或"错误:原因"，不需要其他解释。"""
+问题改写: {rewrite}
+槽位模板：{slot_info_prompt}
+
+填充的槽位信息: {slots}
+
+槽位信息应该准确提取问题中的关键实体和属性，如软件名称、功能名称、错误信息等。请分析这些槽位是否准确填充，并以JSON格式返回结果。请提供一个0到1之间的置信度得分，表示你对判断的确信程度。
+
+{format_instructions}
+"""
 
         try:
             response = llm.invoke(prompt)
-            result = response.content.strip()
+            result = parser.parse(response.content)
-            
+            return result.is_correct, result.reason, result.confidence_score
-            if result.startswith("正确"):
-                return True, ""
-            else:
-                error_reason = result.replace("错误:", "").strip() if "错误:" in result else result
-                return False, error_reason
         except Exception as e:
-            logging.warning(f"验证检索关键词时出错: {e}")
+            logging.warning(f"验证槽位填充时出错: {e}")
-            return False, f"验证过程出错: {str(e)}"
+            return False, f"验证过程出错: {str(e)}", 0.0
+
+    def validate_retrieve_content(self, llm, rewrite, retrieve_content):
+        """
+        验证检索内容是否正确
+        
+        Args:
+            llm: LLM模型
+            rewrite: 问题改写
+            retrieve_content: 检索内容(可能是JSON字符串或文本)
+            
+        Returns:
+            (bool, str, float): 是否正确，错误原因(如果有)，置信度
+        """
+        # 解析检索内容
+        try:
+            if isinstance(retrieve_content, str) and retrieve_content.strip():
+                if retrieve_content.startswith('{') or retrieve_content.startswith('['):
+                    content = json.loads(retrieve_content)
+                else:
+                    content = retrieve_content
+            else:
+                content = retrieve_content
+        except:
+            content = retrieve_content
+        
+        parser = self.create_validation_parser()
+        format_instructions = parser.get_format_instructions()
+        
+        prompt = f"""
+        背景：用户正在使用电力造价软件，提出的问题可能涉及电力造价软件的使用帮助，也可能涉及电力造价专业知识。我针对用户问题检索了相关内容，请评估这些检索内容是否与问题相关、是否准确。
+
+问题改写: {rewrite}
+检索内容: {content}
+
+检索内容应该与问题主题相关，能够提供有用的信息来回答问题。请分析检索内容是否相关、准确，并以JSON格式返回结果。请提供一个0到1之间的置信度得分，表示你对判断的确信程度。
+
+{format_instructions}
+"""
+
+        try:
+            response = llm.invoke(prompt)
+            result = parser.parse(response.content)
+            return result.is_correct, result.reason, result.confidence_score
+        except Exception as e:
+            logging.warning(f"验证检索内容时出错: {e}")
+            return False, f"验证过程出错: {str(e)}", 0.0
 
     def validate_rewrite(self, llm, query, rewrite):
         """
@@ -208,28 +281,29 @@ class ExcelDataValidator:
             rewrite: 问题改写
             
         Returns:
-            (bool, str): 是否正确，错误原因(如果有)
+            (bool, str, float): 是否正确，错误原因(如果有)，置信度
         """
+        parser = self.create_validation_parser()
+        format_instructions = parser.get_format_instructions()
+        
         prompt = f"""
         背景：用户正在使用电力造价软件，提出的问题可能涉及电力造价软件的使用帮助，也可能涉及电力造价专业知识。我对用户问题进行了改写，请评估以下问题改写是否正确。
 
 原始问题: {query}
 问题改写: {rewrite}
 
-问题改写应该保持原问题的核心意图，同时使表达更加清晰、完整。请分析改写是否准确。只需返回"正确"或"错误:原因"，不需要其他解释。"""
+问题改写应该保持原问题的核心意图，同时使表达更加清晰、完整。请分析改写是否准确，并以JSON格式返回结果。请提供一个0到1之间的置信度得分，表示你对判断的确信程度。
+
+{format_instructions}
+"""
 
         try:
             response = llm.invoke(prompt)
-            result = response.content.strip()
+            result = parser.parse(response.content)
-            
+            return result.is_correct, result.reason, result.confidence_score
-            if result.startswith("正确"):
-                return True, ""
-            else:
-                error_reason = result.replace("错误:", "").strip() if "错误:" in result else result
-                return False, error_reason
         except Exception as e:
             logging.warning(f"验证问题改写时出错: {e}")
-            return False, f"验证过程出错: {str(e)}"
+            return False, f"验证过程出错: {str(e)}", 0.0
     
     def validate_row(self, llm, row_data):
         """
@@ -240,42 +314,110 @@ class ExcelDataValidator:
             row_data: (index, row)元组
             
         Returns:
-            (index, is_all_correct, error_phase, error_reason): 行索引，是否全部正确，错误环节，错误原因
+            (index, is_all_correct, error_phase, error_reason, confidence_score): 行索引，是否全部正确，错误环节，错误原因，置信度
         """
         index, row = row_data
-        query = row["提问"]
+        query = row["问题"]
-        query_keys = row["问题拆解"]
+        query_class = row.get("问题分类", "")
-        vertical_class = row["一级分类"]
+        rewrite = row.get("问题改写", "")
-        sub_class = row["二级分类"]
+        slot_info = row.get("槽点信息", "")
-        rewrite = row["问题改写"]
+        retrieve_content = row.get("检索的内容", "")
-        keywords_str = row["检索的关键词"]
+        
+        if self.debug:
+            logging.info(f"开始验证行 {index}:")
+            logging.info(f"  问题: {query}")
+            logging.info(f"  问题分类: {query_class}")
+            logging.info(f"  问题改写: {rewrite}")
         
         try:
-            # 1. 验证问题分类
-            is_correct, error_reason = self.validate_classification(llm, query, vertical_class, sub_class)
-            if not is_correct:
-                return index, False, "问题分类", error_reason
 
-            # 2. 验证问题拆解
+            confidence_score = 0.0
-            is_correct, error_reason = self.validate_query_keys(llm, query, query_keys)
+            # 1. 验证问题改写
-            if not is_correct:
+            if rewrite:
-                return index, False, "问题拆解", error_reason
+                if self.debug:
+                    logging.info(f"  验证问题改写...")
                 
-            # 3. 验证检索关键词
+                result = self.validate_rewrite(llm, query, rewrite)
-            is_correct, error_reason = self.validate_keywords(llm, query, query_keys, keywords_str)
+                if isinstance(result, tuple) and len(result) >= 3:
-            if not is_correct:
+                    is_correct, error_reason, rewrite_confidence = result[:3]
-                return index, False, "关键词检索", error_reason
+                    confidence_score = max(confidence_score, rewrite_confidence)
                     
-            # 4. 验证问题改写
+                    if self.debug:
-            is_correct, error_reason = self.validate_rewrite(llm, query, rewrite)
+                        logging.info(f"  问题改写验证结果: {'通过' if is_correct else '不通过'}, 置信度: {rewrite_confidence:.2f}")
-            if not is_correct:
+                        if not is_correct:
-                return index, False, "问题改写", error_reason
+                            logging.info(f"  错误原因: {error_reason}")
                     
-            return index, True, "", ""
+                    if not is_correct:
+                        return index, False, "问题改写", error_reason, rewrite_confidence
+                    
+            # 2. 验证问题分类
+            if query_class:
+                if self.debug:
+                    logging.info(f"  验证问题分类...")
+                
+                query_class_list = query_class.split(" - ")
+                if len(query_class_list) >= 2:
+                    result = self.validate_classification(llm, rewrite, query_class_list[0], query_class_list[1])
+                    if isinstance(result, tuple) and len(result) >= 3:
+                        is_correct, error_reason, confidence_score = result[:3]
+                        
+                        if self.debug:
+                            logging.info(f"  问题分类验证结果: {'通过' if is_correct else '不通过'}, 置信度: {confidence_score:.2f}")
+                            if not is_correct:
+                                logging.info(f"  错误原因: {error_reason}")
+                        
+                        if not is_correct:
+                            return index, False, "问题分类", error_reason, confidence_score
+            
+
+            
+            # 3. 验证槽位填充
+            if slot_info:
+                if self.debug:
+                    logging.info(f"  验证槽位填充...")
+                
+                result = self.validate_slot(llm, rewrite, slot_info, query_class_list[0], query_class_list[1])
+                if isinstance(result, tuple) and len(result) >= 3:
+                    is_correct, error_reason, slot_confidence = result[:3]
+                    confidence_score = max(confidence_score, slot_confidence)
+                    
+                    if self.debug:
+                        logging.info(f"  槽位填充验证结果: {'通过' if is_correct else '不通过'}, 置信度: {slot_confidence:.2f}")
+                        if not is_correct:
+                            logging.info(f"  错误原因: {error_reason}")
+                    
+                    if not is_correct:
+                        return index, False, "槽位填充", error_reason, slot_confidence
+            
+            # 4. 验证检索内容
+            if retrieve_content:
+                if self.debug:
+                    logging.info(f"  验证检索内容...")
+                
+                result = self.validate_retrieve_content(llm, rewrite, retrieve_content)
+                if isinstance(result, tuple) and len(result) >= 3:
+                    is_correct, error_reason, retrieve_confidence = result[:3]
+                    confidence_score = max(confidence_score, retrieve_confidence)
+                    
+                    if self.debug:
+                        logging.info(f"  检索内容验证结果: {'通过' if is_correct else '不通过'}, 置信度: {retrieve_confidence:.2f}")
+                        if not is_correct:
+                            logging.info(f"  错误原因: {error_reason}")
+                    
+                    if not is_correct:
+                        return index, False, "检索内容", error_reason, retrieve_confidence
+
+            if self.debug:
+                logging.info(f"  行 {index} 验证完成: 通过, 总置信度: {confidence_score:.2f}")
+            
+            return index, True, "", "", confidence_score
         except Exception as e:
             error_msg = f"处理行 {index} 时发生错误: {str(e)}"
             logging.error(error_msg)
-            return index, False, "处理错误", error_msg
+            if self.debug:
+                import traceback
+                logging.error(traceback.format_exc())
+            return index, False, "处理错误", error_msg, 0.0
     
     def process_batch(self, llm, batch_data):
         """处理一批数据"""
@@ -288,7 +430,7 @@ class ExcelDataValidator:
         """创建多个LLM实例"""
         api_key = os.getenv("OPENAI_API_KEY")
         base_url = os.getenv("OPENAI_API_BASE")
-        model_name = os.getenv("LLM_MODEL_NAME", "gpt-3.5-turbo")
+        model_name = "deepseek-ai/DeepSeek-R1"
         
         llm_params = {"temperature": 0.7, "model": model_name}
         if api_key:
@@ -298,7 +440,7 @@ class ExcelDataValidator:
 
         return [OpenAiLLM(**llm_params) for _ in range(count)]
     
-    def validate(self, input_file=None, output_file=None, workers=None, batch_size=None):
+    def validate(self, input_file=None, output_file=None, workers=None, batch_size=None, debug=None):
         """
         执行验证过程
         
@@ -307,6 +449,7 @@ class ExcelDataValidator:
             output_file: 输出结果Excel文件路径
             workers: 并行工作线程数
             batch_size: 每批处理的行数
+            debug: 是否启用调试模式（串行处理）
             
         Returns:
             验证后的DataFrame
@@ -315,6 +458,7 @@ class ExcelDataValidator:
         output_file = output_file or self.output_file
         workers = workers or self.workers
         batch_size = batch_size or self.batch_size
+        debug = debug if debug is not None else self.debug
         
         # 读取数据
         df = self.load_data_from_excel(input_file)
@@ -325,36 +469,64 @@ class ExcelDataValidator:
         df["验证结果"] = ""
         df["错误环节"] = ""
         df["错误原因"] = ""
+        df["置信度"] = 0.0
         
-        # 准备数据批次
+        # 准备数据
         all_rows = list(df.iterrows())
-        batches = [all_rows[i:i+batch_size] for i in range(0, len(all_rows), batch_size)]
         
-        # 创建多个LLM实例
+        # 创建LLM实例
-        llm_instances = self.create_llm_instances(min(workers, len(batches)))
+        llm = self.create_llm_instances(1)[0]
         
-        # 使用线程池处理数据
+        # 根据模式选择处理方式
         all_results = []
-        with concurrent.futures.ThreadPoolExecutor(max_workers=workers) as executor:
+        if debug:
-            # 为每个批次分配一个LLM实例
+            # 调试模式：串行处理
-            future_to_batch = {
+            logging.info("启用调试模式，使用串行处理...")
-                executor.submit(self.process_batch, llm_instances[i % len(llm_instances)], batch): 
+            for i, row_data in enumerate(all_rows):
-                i for i, batch in enumerate(batches)
+                logging.info(f"处理第 {i+1}/{len(all_rows)} 行...")
-            }
+                result = self.validate_row(llm, row_data)
+                all_results.append(result)
+                # 实时更新DataFrame
+                index, is_correct, error_phase, error_reason, confidence_score = result
+                df.at[index, "验证结果"] = "通过" if is_correct else "不通过"
+                df.at[index, "错误环节"] = error_phase
+                df.at[index, "错误原因"] = error_reason
+                df.at[index, "置信度"] = confidence_score
+                # 输出当前结果
+                logging.info(f"行 {index} 验证结果: {'通过' if is_correct else '不通过'}, 错误环节: {error_phase}, 错误原因: {error_reason}, 置信度: {confidence_score:.2f}")
+        else:
+            # 正常模式：并行处理
+            batches = [all_rows[i:i+batch_size] for i in range(0, len(all_rows), batch_size)]
+            llm_instances = self.create_llm_instances(min(workers, len(batches)))
             
-            # 使用tqdm显示进度条
+            with concurrent.futures.ThreadPoolExecutor(max_workers=workers) as executor:
-            for future in tqdm(concurrent.futures.as_completed(future_to_batch), total=len(batches), desc="批次处理进度"):
+                # 为每个批次分配一个LLM实例
-                batch_results = future.result()
+                future_to_batch = {
-                all_results.extend(batch_results)
+                    executor.submit(self.process_batch, llm_instances[i % len(llm_instances)], batch): 
+                    i for i, batch in enumerate(batches)
+                }
+                
+                # 使用tqdm显示进度条
+                for future in tqdm(concurrent.futures.as_completed(future_to_batch), total=len(batches), desc="批次处理进度"):
+                    batch_results = future.result()
+                    all_results.extend(batch_results)
         
         # 按行索引排序结果，确保与原始数据顺序一致
         all_results.sort(key=lambda x: x[0])
         
         # 将结果填充到DataFrame
-        for index, is_correct, error_phase, error_reason in all_results:
+        for result in all_results:
-            df.at[index, "验证结果"] = "通过" if is_correct else "不通过"
+            if len(result) >= 5:
-            df.at[index, "错误环节"] = error_phase
+                index, is_correct, error_phase, error_reason, confidence_score = result
-            df.at[index, "错误原因"] = error_reason
+                df.at[index, "验证结果"] = "通过" if is_correct else "不通过"
+                df.at[index, "错误环节"] = error_phase
+                df.at[index, "错误原因"] = error_reason
+                df.at[index, "置信度"] = confidence_score
+            else:
+                index, is_correct, error_phase, error_reason = result
+                df.at[index, "验证结果"] = "通过" if is_correct else "不通过"
+                df.at[index, "错误环节"] = error_phase
+                df.at[index, "错误原因"] = error_reason
         
         # 保存结果
         if output_file is None:
@@ -381,25 +553,34 @@ class ExcelDataValidator:
         for phase, count in error_stats.items():
             logging.info(f"- {phase}: {count} 条")
 
+    def create_validation_parser(self):
+        """创建验证结果解析器"""
+        return PydanticOutputParser(pydantic_object=ValidationResult)
+
 
 def main():
     """主函数"""
     # 解析命令行参数
-    input_excel = os.path.join(os.path.dirname(__file__), "..", "..", "data", "excel", "问题分类重写结果")
+    input_excel = os.path.join(os.path.dirname(__file__), "..", "..", "data", "excel", "1500条点踩软件问题测试_检索结果.xlsx")
     output_excel = os.path.join(os.path.dirname(__file__), "..", "..", "data", "excel", "自动验证_问题分类重写结果.xlsx")
+
     parser = argparse.ArgumentParser(description="验证Excel数据中的问题分类、问题拆解、检索关键词和问题改写")
-    parser.add_argument("--input", "-i", type=str, required=True, help="输入Excel文件路径", default=input_excel)
+    parser.add_argument("--input", "-i", type=str, help="输入Excel文件路径", default=input_excel)
     parser.add_argument("--output", "-o", type=str, help="输出结果Excel文件路径", default=output_excel)
-    parser.add_argument("--workers", "-w", type=int, default=2, help="并行工作线程数")
+    parser.add_argument("--workers", "-w", type=int, default=20, help="并行工作线程数")
     parser.add_argument("--batch-size", "-b", type=int, default=5, help="每批处理的行数")
+    parser.add_argument("--debug", "-d", action="store_true", help="启用调试模式（串行处理）")
+
     args = parser.parse_args()
+    is_debug = hasattr(sys, 'gettrace') and sys.gettrace() is not None
     
     # 创建验证器实例并执行验证
     validator = ExcelDataValidator(
         input_file=args.input,
         output_file=args.output,
         workers=args.workers,
-        batch_size=args.batch_size
+        batch_size=args.batch_size,
+        debug=is_debug
     )
     validator.validate()
 

rag2_0/demo/vectorize_save_noun.py

@@ -9,6 +9,8 @@ Description: 专业名词向量化和保存的示例程序
 import os
 import json
 from dotenv import load_dotenv
+import sys
+sys.path.append(os.getcwd())
 from rag2_0.intent_recognition import ProfessionalNounVectorizer
 import logging
 

rag2_0/intent_recognition/DataModels.py

@@ -149,7 +149,7 @@ class SlotBase(BaseModel):
 class SoftwareFunctionSlots(SlotBase):
     software_name: str = Field(default="", description="软件名称")
     function_name: str = Field(default="", description="具体功能名称")
-    operation: str = Field(default="", description="用户操作意图(如何使用功能、功能入口、功能使用场景)")
+    operation: str = Field(default="", description="用户操作意图(如何使用功能、功能入口、功能使用场景、是否支持该功能)")
     project_type: Optional[str] = Field(default="单工程", description="工程类型(单工程、多工程、批次工程), 未明确提及则默认下是(单工程)")
     software_version: Optional[str] = Field(default="", description="软件版本")
     operation_steps: Optional[str] = Field(default="", description="操作步骤描述")

rag2_0/tool/APIKeyManager.py

@@ -271,7 +271,7 @@ if __name__ == "__main__":
     instance = APIKeyManager.get_instance()
     stats = instance.get_usage_stats()
     all_balance=0.0
-    buy_balance=16 * 10 * 14 # 购买16次，一次10条api_key，每个api_key有14元
+    buy_balance=17 * 10 * 14 # 购买16次，一次10条api_key，每个api_key有14元
     invalid_api_keys = []
     for key, data in stats.items():
         usage_stats = APIKeyManager.get_key_usage_stats(key)