优化对话转工单处理逻辑，调整LLM参数，增强用户问题和解决方案的提取功能，添加槽位填充支持，提升代码结构和可读性。

2025-05-30 11:10:24 +08:00
parent 05caedc4fa
commit d4ff7b6fad
6 changed files with 469 additions and 123 deletions
@@ -11,12 +11,17 @@ import os
 from langchain_openai import ChatOpenAI
 from langchain.output_parsers import PydanticOutputParser
 import json
-from typing import List, Tuple
+from typing import List, Tuple, Dict, Any, Optional, Union
 import re
-from .PromptTemplates import classification_prompt, query_rewrite_prompt, extract_nouns_prompt, classification_info
-from .DataModels import Classification, QueryRewrite, Term, TermList
+from .PromptTemplates import classification_prompt, query_rewrite_prompt, extract_nouns_prompt, classification_info, slot_filling_prompt
+from .DataModels import (
+    Classification, QueryRewrite, Term, TermList,
+    SoftwareFunction, TroubleShooting, ProfessionalConsulting, 
+    DataProblem, FileExtensionConsulting, SoftwareLock, 
+    InstallationDownload, ProblemDiagnosis
+)
 from .ProfessionalNounVector import ProfessionalNounRetriever
-from rag2_0.tool.ModelTool import XinferenceReRankerModel, OpenAiLLM
+from rag2_0.tool.ModelTool import XinferenceReRankerModel, OpenAiLLM, SiliconFlowReRankerModel


 class IntentRecognizer:
@@ -184,7 +189,7 @@ class IntentRecognizer:
        if len(matched_terms) != 0:
            txts = ["名称:" + term.name + "|" + "同义词:" + ";".join(term.synonymous) + "|" + "描述:" + term.description for term in matched_terms]
            # txts = [term.name for term in matched_terms]
-            xinference_reranker = XinferenceReRankerModel()
+            xinference_reranker = SiliconFlowReRankerModel()
            rerank_results = xinference_reranker.rerank(query, txts, top_k=5)
            matched_terms_list = list(matched_terms)
            matched_terms = [matched_terms_list[result["index"]] for result in rerank_results]
@@ -288,4 +293,136 @@ class IntentRecognizer:
            return classification, TermList(terms=[]), QueryRewrite(rewrite=query),[]

        # rewrite = QueryRewrite(rewrite=query)
-        return classification, keywords_terms, rewrite, query_keys
+        return classification, keywords_terms, rewrite, query_keys
+    
+    def fill_slots(self, query: str, classification: Classification, keywords: TermList) -> Dict[str, Any]:
+        """
+        根据分类结果对问题进行槽位填充
+        
+        Args:
+            query: 用户原始问题
+            classification: 意图分类结果
+            keywords: 匹配的关键词列表
+            
+        Returns:
+            填充后的槽位数据模型
+        """
+        # 根据分类结果选择对应的数据模型
+        slot_model = self._get_slot_model(classification)
+        if not slot_model:
+            return {"error": "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, keywords, slot_model)
+        
+        # 检查必填槽位是否都已填充
+        is_complete, missing_slots = filled_slots.check_required_slots()
+        
+        return {
+            "is_complete": is_complete,
+            "missing_slots": missing_slots,
+            "filled_data": filled_slots.model_dump()
+        }
+    
+    def _get_slot_model(self, classification: Classification) -> Optional[type]:
+        """
+        根据分类结果获取对应的槽位模型类
+        
+        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 SoftwareFunction
+            elif classification.sub_classification == "故障排查":
+                return TroubleShooting
+        
+        # 业务问题
+        elif classification.vertical_classification == "业务问题":
+            if classification.sub_classification == "专业咨询":
+                return ProfessionalConsulting
+            elif classification.sub_classification == "数据问题":
+                return DataProblem
+        
+        # 安装下载注册
+        elif classification.vertical_classification == "安装下载":
+            if classification.sub_classification == "后缀名咨询":
+                return FileExtensionConsulting
+            elif classification.sub_classification == "软件锁类":
+                return SoftwareLock
+            elif classification.sub_classification == "安装下载类":
+                return InstallationDownload
+            elif classification.sub_classification == "问题排查类":
+                return ProblemDiagnosis
+        
+        return None
+    
+    def _fill_slots_with_llm(self, query: str, classification: Classification, keywords: TermList, slot_model_class: type) -> Any:
+        """
+        使用LLM进行槽位填充
+        
+        Args:
+            query: 用户原始问题
+            classification: 意图分类结果
+            keywords: 匹配的关键词列表
+            slot_model_class: 槽位模型类
+            
+        Returns:
+            填充后的槽位数据模型实例
+        """
+        # 准备提示词
+        slot_parser = PydanticOutputParser(pydantic_object=slot_model_class)
+        model_schema = json.dumps(slot_model_class.model_json_schema(), ensure_ascii=False)
+        terms_dict = [term.model_dump() for term in keywords.terms]
+        keywords_str = json.dumps(terms_dict, ensure_ascii=False)
+        
+        formatted_prompt = slot_filling_prompt.format(
+            query=query,
+            vertical_classification=classification.vertical_classification,
+            sub_classification=classification.sub_classification,
+            keywords=keywords_str,
+            model_schema=model_schema,
+            output_format=slot_parser.get_format_instructions()
+        )
+        
+        # 调用LLM
+        response = self.llm.invoke(formatted_prompt, False)
+        
+        try:
+            # 尝试解析LLM响应
+            parsed_output = slot_parser.parse(response.content)
+            return parsed_output
+        except Exception as e:
+            # 如果解析失败，创建一个空的模型实例
+            empty_instance = slot_model_class()
+            return empty_instance
+    
+    def process_query_with_slots(self, query: str) -> Dict[str, Any]:
+        """
+        处理用户问题的完整流程，包括槽位填充
+        
+        Args:
+            query: 用户原始问题
+            
+        Returns:
+            包含分类、关键词、改写和槽位填充结果的字典
+        """
+        # 执行基本处理流程
+        classification, keywords, rewrite, query_keys = self.process_query(query)
+        
+        # 如果是有效分类，进行槽位填充
+        slot_filling_result = {}
+        if classification.vertical_classification not in ["其他", "闲聊"] and classification.sub_classification not in ["其他", "闲聊"]:
+            slot_filling_result = self.fill_slots(rewrite.rewrite, classification, keywords)
+        
+        return {
+            "classification": classification.model_dump(),
+            "keywords": keywords.model_dump(),
+            "rewrite": rewrite.model_dump(),
+            "query_keys": query_keys,
+            "slot_filling": slot_filling_result
+        }