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。

rag2_0/intent_recognition/DataModels.py

@@ -8,7 +8,7 @@ Description: 提取和分类的数据模型
 """
 
 from pydantic import BaseModel, Field
-from typing import List, Optional
+from typing import List, Optional, Dict, Tuple
 
 
 # 定义输出模型
@@ -33,4 +33,138 @@ class Classification(BaseModel):
     sub_classification:str = Field(description="一级分类下的二级分类")
 
 class QueryRewrite(BaseModel):
-    rewrite:str = Field(description="问题改写")
+    rewrite:str = Field(description="问题改写")
+
+# 1. 软件问题
+# 1.1 软件功能
+class SoftwareFunction(BaseModel):
+    software_name: str = Field(description="软件名称")
+    function_name: str = Field(description="具体功能名称")
+    operation: str = Field(description="用户操作意图(如何使用功能、功能入口、功能使用场景)")
+    software_version: Optional[str] = Field(None, description="软件版本")
+    operation_steps: Optional[str] = Field(None, description="操作步骤描述")
+    
+    def check_required_slots(self) -> Tuple[bool, Dict[str, str]]:
+        """检查必填槽位是否都存在"""
+        missing_slots = {}
+        if not self.software_name:
+            missing_slots["software_name"] = SoftwareFunction.model_fields["software_name"].description
+        if not self.function_name:
+            missing_slots["function_name"] = SoftwareFunction.model_fields["function_name"].description
+        if not self.operation:
+            missing_slots["operation"] = SoftwareFunction.model_fields["operation"].description
+        return len(missing_slots) == 0, missing_slots
+
+# 1.2 故障排查
+class TroubleShooting(BaseModel):
+    software_name: str = Field(description="软件名称")
+    function_name: str = Field(description="具体功能名称/操作描述")
+    error_message: str = Field(description="报错信息/异常现象")
+    software_version: Optional[str] = Field(None, description="软件版本")
+    os_version: Optional[str] = Field(None, description="操作系统及版本")
+    reproduction_steps: Optional[str] = Field(None, description="故障重现步骤")
+    
+    def check_required_slots(self) -> Tuple[bool, Dict[str, str]]:
+        """检查必填槽位是否都存在"""
+        missing_slots = {}
+        if not self.software_name:
+            missing_slots["software_name"] = TroubleShooting.model_fields["software_name"].description
+        if not self.function_name:
+            missing_slots["function_name"] = TroubleShooting.model_fields["function_name"].description
+        if not self.error_message:
+            missing_slots["error_message"] = TroubleShooting.model_fields["error_message"].description
+        return len(missing_slots) == 0, missing_slots
+
+# 2. 业务问题
+# 2.1 专业咨询
+class ProfessionalConsulting(BaseModel):
+    scene_subject: str = Field(description="场景主体")
+    business_scene: str = Field(description="业务场景描述")
+    software_name: Optional[str] = Field(None, description="软件名称")
+    
+    def check_required_slots(self) -> Tuple[bool, Dict[str, str]]:
+        """检查必填槽位是否都存在"""
+        missing_slots = {}
+        if not self.scene_subject:
+            missing_slots["scene_subject"] = ProfessionalConsulting.model_fields["scene_subject"].description
+        if not self.business_scene:
+            missing_slots["business_scene"] = ProfessionalConsulting.model_fields["business_scene"].description
+        return len(missing_slots) == 0, missing_slots
+
+# 2.2 数据问题
+class DataProblem(BaseModel):
+    expense_type: str = Field(description="费用类型")
+    operation_purpose: str = Field(description="操作目的")
+    software_name: Optional[str] = Field(None, description="软件名称")
+    project_type: Optional[str] = Field(None, description="工程类型")
+    
+    def check_required_slots(self) -> Tuple[bool, Dict[str, str]]:
+        """检查必填槽位是否都存在"""
+        missing_slots = {}
+        if not self.expense_type:
+            missing_slots["expense_type"] = DataProblem.model_fields["expense_type"].description
+        if not self.operation_purpose:
+            missing_slots["operation_purpose"] = DataProblem.model_fields["operation_purpose"].description
+        return len(missing_slots) == 0, missing_slots
+
+# 3. 安装下载注册
+# 3.1 后缀名咨询
+class FileExtensionConsulting(BaseModel):
+    file_extension: str = Field(description="文件后缀名")
+    operation_purpose: str = Field(description="操作目的")
+    file_source: Optional[str] = Field(None, description="文件来源场景")
+    related_software: Optional[str] = Field(None, description="相关软件名称")
+    
+    def check_required_slots(self) -> Tuple[bool, Dict[str, str]]:
+        """检查必填槽位是否都存在"""
+        missing_slots = {}
+        if not self.file_extension:
+            missing_slots["file_extension"] = FileExtensionConsulting.model_fields["file_extension"].description
+        if not self.operation_purpose:
+            missing_slots["operation_purpose"] = FileExtensionConsulting.model_fields["operation_purpose"].description
+        return len(missing_slots) == 0, missing_slots
+
+# 3.2 软件锁类
+class SoftwareLock(BaseModel):
+    lock_type: str = Field(description="锁类型")
+    operation_purpose: str = Field(description="操作目的")
+    lock_number: Optional[str] = Field(None, description="软件锁编号/注册号")
+    
+    def check_required_slots(self) -> Tuple[bool, Dict[str, str]]:
+        """检查必填槽位是否都存在"""
+        missing_slots = {}
+        if not self.lock_type:
+            missing_slots["lock_type"] = SoftwareLock.model_fields["lock_type"].description
+        if not self.operation_purpose:
+            missing_slots["operation_purpose"] = SoftwareLock.model_fields["operation_purpose"].description
+        return len(missing_slots) == 0, missing_slots
+
+# 3.3 安装下载类
+class InstallationDownload(BaseModel):
+    software_name: str = Field(description="软件/插件名称")
+    operation_stage: str = Field(description="操作阶段")
+    os_version: Optional[str] = Field(None, description="操作系统版本")
+    package_source: Optional[str] = Field(None, description="安装包来源/版本号")
+    
+    def check_required_slots(self) -> Tuple[bool, Dict[str, str]]:
+        """检查必填槽位是否都存在"""
+        missing_slots = {}
+        if not self.software_name:
+            missing_slots["software_name"] = InstallationDownload.model_fields["software_name"].description
+        if not self.operation_stage:
+            missing_slots["operation_stage"] = InstallationDownload.model_fields["operation_stage"].description
+        return len(missing_slots) == 0, missing_slots
+
+# 3.4 问题排查类
+class ProblemDiagnosis(BaseModel):
+    error_message: str = Field(description="报错信息/异常现象")
+    software_name: Optional[str] = Field(None, description="软件名称")
+    os_version: Optional[str] = Field(None, description="操作系统版本")
+    
+    def check_required_slots(self) -> Tuple[bool, Dict[str, str]]:
+        """检查必填槽位是否都存在"""
+        missing_slots = {}
+        if not self.error_message:
+            missing_slots["error_message"] = ProblemDiagnosis.model_fields["error_message"].description
+        return len(missing_slots) == 0, missing_slots
+

rag2_0/intent_recognition/IntentRecognition.py

@@ -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
+        }

rag2_0/intent_recognition/PromptTemplates.py

@@ -38,7 +38,7 @@ classification_info="""【垂直领域分类】：
 
 【业务问题包括以下两类】：
 1. 专业咨询：涉及电力造价规范、工程计价规则问题、行业标准解读等
-2. 数据问题：涉及电力造价费用、造价指标等
+2. 数据问题：涉及电力造价费用、造价指标的计算或构成等
 
 【安装下载注册包括以下三类】：
 1. 后缀名咨询：所有涉及文件扩展名的使用场景、软件关联等问题，包括但不限于：询问文件是否由特定软件打开、扩展名与软件的匹配关系、扩展名含义及关联等
@@ -94,7 +94,7 @@ query_rewrite_prompt = """
    b. 执行结构优化：
       - 采用【术语标记】规范标注关键概念
       - 构建主谓宾明确的问题句式
-      - 保持原问题时态与语态特征
+      - 保持原问题时态与语态特征, 保留5W2H问题特征
       - 执行同义词替换：将synonymous中的同义词替换为对应name字段的标准术语
 
 # 输出规范
@@ -132,4 +132,43 @@ query_rewrite_prompt = """
 4. 异常处理机制
    - 当关键词与问题无明显关联时，触发直通输出规则
    - 出现术语冲突时优先保留原始表述
-    """
+    """
+
+slot_filling_prompt = """
+你是一个专业的电力造价领域问题槽位填充助手。你需要从用户问题中提取关键信息，并填充到对应的数据结构中。
+
+【用户问题】
+{query}
+
+【问题分类】
+垂直领域分类: {vertical_classification}
+子分类: {sub_classification}
+
+【已识别关键词】
+{keywords}
+
+【目标数据结构】
+{model_schema}
+
+【输出格式】
+{output_format}
+
+【任务要求】
+1. 仔细分析用户问题，从中提取所有可能的槽位信息
+2. 对于必填槽位，必须尽力从问题中提取，如果确实无法提取则留空
+3. 对于选填槽位，如果能从问题中提取则填写，否则留空
+4. 只输出符合格式的JSON数据，不要有任何额外的解释
+
+【示例】
+用户问题: "我的西藏Z1软件安装后闪退，提示缺少组件，怎么解决？"
+分类: 软件问题/故障排查
+输出:
+{{
+  "software_name": "西藏Z1软件",
+  "function_name": "软件安装",
+  "error_message": "闪退，提示缺少组件",
+  "software_version": null,
+  "os_version": null,
+  "reproduction_steps": "软件安装后"
+}}
+"""