ouyangyouzhang/QueryRewrite
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

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。

Browse Source
This commit is contained in:
ouyangyouzhang
2025-07-21 10:11:34 +08:00
parent 0dda581c8e
commit 17ff99bad1
1 changed files with 109 additions and 91 deletions
Download Patch File Download Diff File Expand all files Collapse all files
rag2_0/demo/dialogue_to_workorder.py
+109 -91
View File
@@ -135,15 +135,23 @@ class DialogueToWorkorder:
self.product_line_parser = PydanticOutputParser(pydantic_object=ProductLine) self.product_line_parser = PydanticOutputParser(pydantic_object=ProductLine)
# 初始化LLM模型 # 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 { self.llm_params = llm_params or {
"temperature": 0.2, "temperature": 0.2,
"top_p":0.95, "top_p":0.95,
"model": "deepseek-ai/DeepSeek-R1", "model": "deepseek-r1",
"api_key": os.getenv("OPENAI_API_KEY"), "api_key": "25t%Syu6I9yxX2IuTN",
"base_url": os.getenv("OPENAI_API_BASE"), "base_url": "http://10.1.0.154:8000/v1",
"timeout": httpx.Timeout(600.0) "timeout": httpx.Timeout(600.0)
} }
self.llm = self._get_llm_instance() self.llm = self._get_llm_instance()
def _get_llm_instance(self): 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) 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。 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} {output_format}
输出示例： 输出示例：
@@ -283,15 +291,15 @@ class DialogueToWorkorder:
output_format = self.user_question_and_solution_parser.get_format_instructions() output_format = self.user_question_and_solution_parser.get_format_instructions()
llm_prompt = prompt.format(output_format=output_format, dialogue_str=dialogue_str) 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: try:
if response.content.count('user_question') == 1: if clean_output.count('user_question') == 1:
user_question_and_solution = self.user_question_and_solution_parser.parse(response.content) user_question_and_solution = self.user_question_and_solution_parser.parse(clean_output)
return [user_question_and_solution] return [user_question_and_solution]
else: else:
array_pattern = r'\[\s*(\{[\s\S]*?\}(?:\s*,\s*\{[\s\S]*?\})*)\s*\]' 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: if array_match:
# 找到了JSON数组 # 找到了JSON数组
json_array_str = '[' + array_match.group(1) + ']' json_array_str = '[' + array_match.group(1) + ']'
@@ -313,8 +321,9 @@ class DialogueToWorkorder:
except Exception as e: except Exception as e:
output_format = self.user_question_and_solution_list_parser.get_format_instructions() output_format = self.user_question_and_solution_list_parser.get_format_instructions()
llm_prompt = prompt.format(output_format=output_format, dialogue_str=dialogue_str) 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)
user_question_and_solution_temp = self.user_question_and_solution_list_parser.parse(response.content) 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_temp.user_question_list
return [user_question_and_solution] return [user_question_and_solution]
@@ -345,8 +354,9 @@ class DialogueToWorkorder:
{dialogue_str} {dialogue_str}
""" """
response = self.llm.invoke(user_prompt=prompt, need_retry=False) response = self.llm.invoke(user_prompt=prompt, need_retry=False, api_key=self.api_key)
product_name_and_module_name = self.product_name_and_module_name_parser.parse(response.content) 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 return product_name_and_module_name.product_name, product_name_and_module_name.module_name
@@ -374,8 +384,9 @@ class DialogueToWorkorder:
{dialogue_str} {dialogue_str}
""" """
response = self.llm.invoke(user_prompt=prompt, need_retry=False) response = self.llm.invoke(user_prompt=prompt, need_retry=False, api_key=self.api_key)
product_line = self.product_line_parser.parse(response.content) 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 return product_line.product_line
@@ -410,8 +421,9 @@ class DialogueToWorkorder:
{dialogue_str} {dialogue_str}
""" """
response = self.llm.invoke(user_prompt=prompt, need_retry=False) response = self.llm.invoke(user_prompt=prompt, need_retry=False, api_key=self.api_key)
question_type = self.question_type_parser.parse(response.content) 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 return question_type.question_type
@@ -446,8 +458,9 @@ class DialogueToWorkorder:
""" """
response = self.llm.invoke(user_prompt=prompt, need_retry=False) response = self.llm.invoke(user_prompt=prompt, need_retry=False, api_key=self.api_key)
is_complaint = self.is_complaint_parser.parse(response.content) 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, return (is_complaint.is_dissatisfaction,
is_complaint.dissatisfaction_level, is_complaint.dissatisfaction_level,
@@ -458,81 +471,85 @@ class DialogueToWorkorder:
"""处理单个会话的函数，用于多线程并发""" """处理单个会话的函数，用于多线程并发"""
# if conversation_id!="b157aa91-3acb-11f0-a191-4fb224ef4b40": # if conversation_id!="b157aa91-3acb-11f0-a191-4fb224ef4b40":
# return [] # return []
# 获取工单基本信息 try:
base_workorder_dict = self.get_workorder_dict(conversation_rows) # 获取工单基本信息
# 分析用户问题和解决方案 base_workorder_dict = self.get_workorder_dict(conversation_rows)
user_question_list = self.get_user_question_and_solution(conversation_rows) # 分析用户问题和解决方案
user_question_list = self.get_user_question_and_solution(conversation_rows)
user_question_str="" user_question_str=""
for user_question in user_question_list: for user_question in user_question_list:
user_question_str = user_question_str + user_question.user_question.strip() + "\n" user_question_str = user_question_str + user_question.user_question.strip() + "\n"
user_question_str = user_question_str.strip() user_question_str = user_question_str.strip()
solution_str="" solution_str=""
for user_question in user_question_list: for user_question in user_question_list:
solution_str = solution_str + user_question.solution.strip() + "\n" solution_str = solution_str + user_question.solution.strip() + "\n"
solution_str = solution_str.strip() solution_str = solution_str.strip()
# 分析是否抱怨、是否投诉、抱怨级别 # 分析是否抱怨、是否投诉、抱怨级别
is_dissatisfaction, dissatisfaction_level, dissatisfaction_reasoning, is_complaint = ( is_dissatisfaction, dissatisfaction_level, dissatisfaction_reasoning, is_complaint = (
self.get_is_complaint_and_is_complaint_level(conversation_rows)) self.get_is_complaint_and_is_complaint_level(conversation_rows))
# 分析问题类型 # 分析问题类型
problem_type = self.get_problem_type(conversation_rows, user_question_str, solution_str) problem_type = self.get_problem_type(conversation_rows, user_question_str, solution_str)
# 分析产品线 # 分析产品线
product_line = self.get_product_line(conversation_rows, product_detail_dict, user_question_str, solution_str) product_line = self.get_product_line(conversation_rows, product_detail_dict, user_question_str, solution_str)
# 分析产品名称和模块名称 # 分析产品名称和模块名称
if product_line != '': if product_line != '':
product_name, module_name = self.get_product_name_and_module_name( product_name, module_name = self.get_product_name_and_module_name(
product_line, conversation_rows, product_detail_dict, user_question_str, solution_str) product_line, conversation_rows, product_detail_dict, user_question_str, solution_str)
else: else:
product_name = '' product_name = ''
module_name = '' module_name = ''
# 创建工单列表 # 创建工单列表
workorder_list = [] workorder_list = []
# 更新工单字典
# base_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',
# "抱怨内容": dissatisfaction_reasoning if is_dissatisfaction else '',
# "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
# })
# 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
# 创建新的工单字典，复制基本信息
workorder_dict = base_workorder_dict.copy()
# 更新工单字典 # 更新工单字典
workorder_dict.update({ # base_workorder_dict.update({
"产品线": product_line, # "产品线": product_line,
"产品名称": product_name, # "产品名称": product_name,
"模块名称": module_name, # "模块名称": module_name,
"客户问题": user_question_str, # "客户问题": user_question_str,
"问题类型": problem_type, # "问题类型": problem_type,
"是否抱怨": "是" if is_dissatisfaction else '否', # "是否抱怨": "是" if is_dissatisfaction else '否',
"抱怨内容": dissatisfaction_reasoning if is_dissatisfaction else '', # "抱怨内容": dissatisfaction_reasoning if is_dissatisfaction else '',
"抱怨级别": dissatisfaction_level if is_dissatisfaction else '', # "抱怨级别": dissatisfaction_level if is_dissatisfaction else '',
"是否投诉": "是" if is_complaint else '否', # "是否投诉": "是" if is_complaint else '否',
"解决方案": solution_str # "解决方案": 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
# 创建新的工单字典，复制基本信息
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',
"抱怨内容": dissatisfaction_reasoning if is_dissatisfaction else '',
"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
})
# 将工单添加到列表中
workorder_list.append(workorder_dict)
# 将工单添加到列表中 return workorder_list
workorder_list.append(workorder_dict) except Exception as e:
logger.error(f"处理会话ID: {conversation_id} 时发生错误: {e}")
return workorder_list return []
def analyze_conversation_data(self, conversation_excel_path, product_detail_excel_path, max_workers=10, start_date=None, end_date=None): 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""" """分析会话数据主流程，使用多线程并发处理"""
@@ -588,9 +605,10 @@ class DialogueToWorkorder:
conversation_id = future_to_conversation[future] conversation_id = future_to_conversation[future]
try: try:
result_workorders = future.result() result_workorders = future.result()
# 将每个会话的所有工单添加到总列表中 if result_workorders:
workorder_dict_list.extend(result_workorders) # 将每个会话的所有工单添加到总列表中
logger.info(f"完成处理会话ID: {conversation_id}，生成工单数量: {len(result_workorders)}") 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)}")
except Exception as exc: except Exception as exc:
logger.error(f"处理会话ID: {conversation_id} 时发生错误: {exc}") logger.error(f"处理会话ID: {conversation_id} 时发生错误: {exc}")
@@ -678,7 +696,7 @@ def parse_arguments():
help='会话内容Excel文件路径') help='会话内容Excel文件路径')
parser.add_argument('--product_detail_file', type=str, required=False, parser.add_argument('--product_detail_file', type=str, required=False,
help='产品详情Excel文件路径') 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') 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", 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') help='开始日期，格式为YYYY-MM-DD')