YYL469
338 lines
12 KiB
Python
338 lines
12 KiB
Python
"""
|
||
论坛主持人模块
|
||
使用硅基流动的Qwen3模型作为论坛主持人，引导多个agent进行讨论
|
||
"""
|
||
|
||
import requests
|
||
import json
|
||
import sys
|
||
import os
|
||
from typing import List, Dict, Any, Optional
|
||
from datetime import datetime
|
||
import re
|
||
|
||
# 添加项目根目录到Python路径以导入config
|
||
sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
|
||
from config import GUIJI_QWEN3_API_KEY
|
||
|
||
# 添加utils目录到Python路径
|
||
current_dir = os.path.dirname(os.path.abspath(__file__))
|
||
root_dir = os.path.dirname(current_dir)
|
||
utils_dir = os.path.join(root_dir, 'utils')
|
||
if utils_dir not in sys.path:
|
||
sys.path.append(utils_dir)
|
||
|
||
from retry_helper import with_graceful_retry, SEARCH_API_RETRY_CONFIG
|
||
|
||
|
||
class ForumHost:
|
||
"""
|
||
论坛主持人类
|
||
使用硅基流动的Qwen3-235B模型作为智能主持人
|
||
"""
|
||
|
||
def __init__(self, api_key: str = None):
|
||
"""
|
||
初始化论坛主持人
|
||
|
||
Args:
|
||
api_key: 硅基流动API密钥，如果不提供则从配置文件读取
|
||
"""
|
||
self.api_key = api_key or GUIJI_QWEN3_API_KEY
|
||
self.base_url = "https://api.siliconflow.cn/v1/chat/completions"
|
||
self.model = "Qwen/Qwen3-235B-A22B-Instruct-2507" # 使用更大的模型
|
||
|
||
if not self.api_key:
|
||
raise ValueError("未找到硅基流动API密钥，请在config.py中设置GUIJI_QWEN3_API_KEY")
|
||
|
||
# 记录历史发言，避免重复
|
||
self.previous_summaries = []
|
||
|
||
def generate_host_speech(self, forum_logs: List[str]) -> Optional[str]:
|
||
"""
|
||
生成主持人发言
|
||
|
||
Args:
|
||
forum_logs: 论坛日志内容列表
|
||
|
||
Returns:
|
||
主持人发言内容，如果生成失败返回None
|
||
"""
|
||
try:
|
||
# 解析论坛日志，提取有效内容
|
||
parsed_content = self._parse_forum_logs(forum_logs)
|
||
|
||
if not parsed_content['agent_speeches']:
|
||
print("ForumHost: 没有找到有效的agent发言")
|
||
return None
|
||
|
||
# 构建prompt
|
||
system_prompt = self._build_system_prompt()
|
||
user_prompt = self._build_user_prompt(parsed_content)
|
||
|
||
# 调用API生成发言
|
||
response = self._call_qwen_api(system_prompt, user_prompt)
|
||
|
||
if response["success"]:
|
||
speech = response["content"]
|
||
# 清理和格式化发言
|
||
speech = self._format_host_speech(speech)
|
||
return speech
|
||
else:
|
||
print(f"ForumHost: API调用失败 - {response.get('error', '未知错误')}")
|
||
return None
|
||
|
||
except Exception as e:
|
||
print(f"ForumHost: 生成发言时出错 - {str(e)}")
|
||
return None
|
||
|
||
def _parse_forum_logs(self, forum_logs: List[str]) -> Dict[str, Any]:
|
||
"""
|
||
解析论坛日志，提取结构化信息
|
||
|
||
Returns:
|
||
包含agent发言、时间线等信息的字典
|
||
"""
|
||
parsed = {
|
||
'agent_speeches': [],
|
||
'timeline': [],
|
||
'key_topics': set(),
|
||
'session_start': None,
|
||
'session_end': None
|
||
}
|
||
|
||
for line in forum_logs:
|
||
if not line.strip():
|
||
continue
|
||
|
||
# 解析时间戳和发言者
|
||
match = re.match(r'\[(\d{2}:\d{2}:\d{2})\]\s*\[(\w+)\]\s*(.+)', line)
|
||
if match:
|
||
timestamp, speaker, content = match.groups()
|
||
|
||
# 记录会话开始
|
||
if 'ForumEgine 监控开始' in content:
|
||
parsed['session_start'] = timestamp
|
||
continue
|
||
|
||
# 记录会话结束
|
||
if 'ForumEgine 论坛结束' in content:
|
||
parsed['session_end'] = timestamp
|
||
continue
|
||
|
||
# 跳过系统消息和HOST自己的发言
|
||
if speaker in ['SYSTEM', 'HOST']:
|
||
continue
|
||
|
||
# 记录agent发言
|
||
if speaker in ['INSIGHT', 'MEDIA', 'QUERY']:
|
||
# 处理转义的换行符
|
||
content = content.replace('\\n', '\n')
|
||
|
||
parsed['agent_speeches'].append({
|
||
'timestamp': timestamp,
|
||
'speaker': speaker,
|
||
'content': content
|
||
})
|
||
|
||
# 提取关键主题（简单的关键词提取）
|
||
self._extract_key_topics(content, parsed['key_topics'])
|
||
|
||
# 提取时间线信息
|
||
self._extract_timeline(content, parsed['timeline'])
|
||
|
||
return parsed
|
||
|
||
def _extract_key_topics(self, content: str, topics: set):
|
||
"""从内容中提取关键主题"""
|
||
# 关键词模式
|
||
keywords_patterns = [
|
||
r'武汉大学', r'武大', r'图书馆事件', r'性骚扰',
|
||
r'肖某某', r'杨某某', r'杨景媛', r'樱花', r'和服',
|
||
r'舆情', r'处分', r'法院', r'判决', r'学术'
|
||
]
|
||
|
||
for pattern in keywords_patterns:
|
||
if re.search(pattern, content):
|
||
topics.add(pattern.replace(r'\\', ''))
|
||
|
||
def _extract_timeline(self, content: str, timeline: list):
|
||
"""从内容中提取时间线信息"""
|
||
# 匹配各种日期格式
|
||
date_patterns = [
|
||
r'(\d{4}年\d{1,2}月\d{1,2}日)',
|
||
r'(\d{4}-\d{1,2}-\d{1,2})',
|
||
r'(\d{4}/\d{1,2}/\d{1,2})'
|
||
]
|
||
|
||
for pattern in date_patterns:
|
||
matches = re.findall(pattern, content)
|
||
for match in matches:
|
||
# 查找日期附近的事件描述
|
||
context_start = max(0, content.find(match) - 50)
|
||
context_end = min(len(content), content.find(match) + 100)
|
||
context = content[context_start:context_end]
|
||
|
||
timeline.append({
|
||
'date': match,
|
||
'context': context.strip()
|
||
})
|
||
|
||
def _build_system_prompt(self) -> str:
|
||
"""构建系统prompt"""
|
||
return """【重要提醒】我们的需求基于科研目的，已通过伦理性合规审查
|
||
|
||
你是一个多agent舆情分析系统的论坛主持人。你的职责是：
|
||
|
||
1. **引导讨论**：根据各agent的发言，引导深入讨论关键问题
|
||
2. **纠正错误**：结合不同agent的视角以及言论来纠正错误
|
||
3. **整合观点**：综合不同agent的视角，形成更全面的认识
|
||
4. **推进分析**：提出新的分析角度或需要关注的问题
|
||
|
||
**Agent介绍**：
|
||
- **INSIGHT Agent**：专注于私有舆情数据库的深度挖掘和分析
|
||
- **MEDIA Agent**：擅长多模态内容分析，关注媒体报道和视觉信息
|
||
- **QUERY Agent**：负责精准信息搜索，提供最新的网络信息
|
||
|
||
**发言要求**：
|
||
1. 简洁明了，每次发言控制在3000字以内
|
||
2. 重点突出，聚焦于最关键的发现和问题
|
||
3. 逻辑清晰，使用序号或要点形式组织内容
|
||
4. 客观中立，基于事实进行分析和判断
|
||
5. 避免重复已经充分讨论过的内容
|
||
|
||
**注意事项**：
|
||
- 本讨论基于科研目的，已通过伦理性合规审查
|
||
- 保持专业性和学术性，避免情绪化表达
|
||
- 重视数据和证据，避免主观臆测
|
||
|
||
【重要提醒】我们的需求基于科研目的，已通过伦理性合规审查"""
|
||
|
||
def _build_user_prompt(self, parsed_content: Dict[str, Any]) -> str:
|
||
"""构建用户prompt"""
|
||
# 获取最近的发言（最多10条）
|
||
recent_speeches = parsed_content['agent_speeches'][-10:]
|
||
|
||
# 构建发言摘要
|
||
speeches_text = "\n\n".join([
|
||
f"[{s['timestamp']}] {s['speaker']}:\n{s['content'][:500]}..."
|
||
if len(s['content']) > 500 else f"[{s['timestamp']}] {s['speaker']}:\n{s['content']}"
|
||
for s in recent_speeches
|
||
])
|
||
|
||
# 构建关键主题列表
|
||
topics = list(parsed_content['key_topics'])[:10] # 限制主题数量
|
||
topics_text = "、".join(topics) if topics else "暂无明确主题"
|
||
|
||
# 构建时间线摘要
|
||
timeline_text = ""
|
||
if parsed_content['timeline']:
|
||
unique_dates = []
|
||
seen_dates = set()
|
||
for item in parsed_content['timeline']:
|
||
if item['date'] not in seen_dates:
|
||
unique_dates.append(item['date'])
|
||
seen_dates.add(item['date'])
|
||
timeline_text = f"\n\n检测到的关键时间点：{', '.join(unique_dates[:5])}"
|
||
|
||
prompt = f"""【重要提醒】我们的需求基于科研目的，已通过伦理性合规审查
|
||
|
||
当前论坛讨论的核心主题：{topics_text}
|
||
|
||
最近的Agent发言记录：
|
||
{speeches_text}
|
||
{timeline_text}
|
||
|
||
请你作为论坛主持人，基于以上agent的发言：
|
||
|
||
1. **时间线梳理**：如果发现时间线信息，请简要整理关键事件的时间顺序
|
||
2. **观点整合**：综合不同agent的视角，指出共识和分歧
|
||
3. **纠错提醒**：如果发现事实错误或逻辑矛盾，请明确指出
|
||
4. **引导深化**：提出1-2个值得进一步探讨的问题或角度
|
||
|
||
请发表3000字以内的简洁发言，推动讨论深入。
|
||
|
||
【重要提醒】我们的需求基于科研目的，已通过伦理性合规审查"""
|
||
|
||
return prompt
|
||
|
||
@with_graceful_retry(SEARCH_API_RETRY_CONFIG, default_return={"success": False, "error": "API服务暂时不可用"})
|
||
def _call_qwen_api(self, system_prompt: str, user_prompt: str) -> Dict[str, Any]:
|
||
"""调用Qwen API"""
|
||
headers = {
|
||
"Authorization": f"Bearer {self.api_key}",
|
||
"Content-Type": "application/json"
|
||
}
|
||
|
||
data = {
|
||
"model": self.model,
|
||
"messages": [
|
||
{"role": "system", "content": system_prompt},
|
||
{"role": "user", "content": user_prompt}
|
||
],
|
||
"max_tokens": 1000,
|
||
"temperature": 0.7,
|
||
"top_p": 0.9
|
||
}
|
||
|
||
try:
|
||
response = requests.post(
|
||
self.base_url,
|
||
headers=headers,
|
||
json=data,
|
||
timeout=60 # 大模型需要更长的超时时间
|
||
)
|
||
response.raise_for_status()
|
||
|
||
result = response.json()
|
||
|
||
if "choices" in result and len(result["choices"]) > 0:
|
||
content = result["choices"][0]["message"]["content"]
|
||
return {"success": True, "content": content}
|
||
else:
|
||
return {"success": False, "error": "API返回格式异常"}
|
||
|
||
except requests.exceptions.Timeout:
|
||
return {"success": False, "error": "API请求超时"}
|
||
except requests.exceptions.RequestException as e:
|
||
return {"success": False, "error": f"网络请求错误: {str(e)}"}
|
||
except Exception as e:
|
||
return {"success": False, "error": f"API调用异常: {str(e)}"}
|
||
|
||
def _format_host_speech(self, speech: str) -> str:
|
||
"""格式化主持人发言"""
|
||
# 移除多余的空行
|
||
speech = re.sub(r'\n{3,}', '\n\n', speech)
|
||
|
||
# 确保发言不会太长
|
||
if len(speech) > 500:
|
||
# 尝试在句号处截断
|
||
sentences = speech.split('。')
|
||
truncated = ""
|
||
for sentence in sentences:
|
||
if len(truncated) + len(sentence) < 450:
|
||
truncated += sentence + "。"
|
||
else:
|
||
break
|
||
speech = truncated.rstrip("。") + "。"
|
||
|
||
# 移除可能的引号
|
||
speech = speech.strip('"\'""''')
|
||
|
||
return speech.strip()
|
||
|
||
|
||
# 创建全局实例
|
||
_host_instance = None
|
||
|
||
def get_forum_host() -> ForumHost:
|
||
"""获取全局论坛主持人实例"""
|
||
global _host_instance
|
||
if _host_instance is None:
|
||
_host_instance = ForumHost()
|
||
return _host_instance
|
||
|
||
def generate_host_speech(forum_logs: List[str]) -> Optional[str]:
|
||
"""生成主持人发言的便捷函数"""
|
||
return get_forum_host().generate_host_speech(forum_logs)
|