mirror of
https://github.com/Nezumi-2711/astrbot_plugin_qq_group_daily_analysis.git
synced 2026-09-22 13:38:43 +00:00
[refactor] (llm_analyzer) 为 llm_analyzer 解耦化
[fix] (LLM 输出提取和增强) 补充原本在话题分析存在但是不存在于 用户头衔和群圣经的 json 提取逻辑
This commit is contained in:
@@ -0,0 +1,5 @@
|
||||
|
||||
.kilocode/
|
||||
.kiro/
|
||||
.vscode/
|
||||
src/analysis/ARCHITECTURE.md
|
||||
@@ -0,0 +1,16 @@
|
||||
"""
|
||||
分析器模块
|
||||
包含各种LLM分析功能的实现
|
||||
"""
|
||||
|
||||
from .base_analyzer import BaseAnalyzer
|
||||
from .topic_analyzer import TopicAnalyzer
|
||||
from .user_title_analyzer import UserTitleAnalyzer
|
||||
from .golden_quote_analyzer import GoldenQuoteAnalyzer
|
||||
|
||||
__all__ = [
|
||||
'BaseAnalyzer',
|
||||
'TopicAnalyzer',
|
||||
'UserTitleAnalyzer',
|
||||
'GoldenQuoteAnalyzer'
|
||||
]
|
||||
@@ -0,0 +1,177 @@
|
||||
"""
|
||||
基础分析器抽象类
|
||||
定义通用分析流程和接口
|
||||
"""
|
||||
|
||||
from abc import ABC, abstractmethod
|
||||
from typing import List, Dict, Tuple, Any, Optional
|
||||
from datetime import datetime
|
||||
from astrbot.api import logger
|
||||
from ...models.data_models import TokenUsage
|
||||
from ..utils.json_utils import parse_json_response
|
||||
from ..utils.llm_utils import call_provider_with_retry, extract_token_usage, extract_response_text
|
||||
import re
|
||||
|
||||
class BaseAnalyzer(ABC):
|
||||
"""
|
||||
基础分析器抽象类
|
||||
定义所有分析器的通用接口和流程
|
||||
"""
|
||||
|
||||
def __init__(self, context, config_manager):
|
||||
"""
|
||||
初始化基础分析器
|
||||
|
||||
Args:
|
||||
context: AstrBot上下文对象
|
||||
config_manager: 配置管理器
|
||||
"""
|
||||
self.context = context
|
||||
self.config_manager = config_manager
|
||||
|
||||
@abstractmethod
|
||||
def get_data_type(self) -> str:
|
||||
"""
|
||||
获取数据类型标识
|
||||
|
||||
Returns:
|
||||
数据类型字符串
|
||||
"""
|
||||
pass
|
||||
|
||||
@abstractmethod
|
||||
def get_max_count(self) -> int:
|
||||
"""
|
||||
获取最大提取数量
|
||||
|
||||
Returns:
|
||||
最大数量
|
||||
"""
|
||||
pass
|
||||
|
||||
@abstractmethod
|
||||
def build_prompt(self, data: Any) -> str:
|
||||
"""
|
||||
构建LLM提示词
|
||||
|
||||
Args:
|
||||
data: 输入数据
|
||||
|
||||
Returns:
|
||||
提示词字符串
|
||||
"""
|
||||
pass
|
||||
|
||||
@abstractmethod
|
||||
def extract_with_regex(self, result_text: str, max_count: int) -> List[Dict]:
|
||||
"""
|
||||
使用正则表达式提取数据
|
||||
|
||||
Args:
|
||||
result_text: LLM响应文本
|
||||
max_count: 最大提取数量
|
||||
|
||||
Returns:
|
||||
提取到的数据列表
|
||||
"""
|
||||
pass
|
||||
|
||||
@abstractmethod
|
||||
def create_data_objects(self, data_list: List[Dict]) -> List[Any]:
|
||||
"""
|
||||
创建数据对象列表
|
||||
|
||||
Args:
|
||||
data_list: 原始数据列表
|
||||
|
||||
Returns:
|
||||
数据对象列表
|
||||
"""
|
||||
pass
|
||||
|
||||
async def analyze(self, data: Any, umo: str = None) -> Tuple[List[Any], TokenUsage]:
|
||||
"""
|
||||
统一的分析流程
|
||||
|
||||
Args:
|
||||
data: 输入数据
|
||||
umo: 模型唯一标识符
|
||||
|
||||
Returns:
|
||||
(分析结果列表, Token使用统计)
|
||||
"""
|
||||
try:
|
||||
# 1. 构建提示词
|
||||
prompt = self.build_prompt(data)
|
||||
logger.info(f"开始{self.get_data_type()}分析,构建提示词完成")
|
||||
|
||||
# 2. 调用LLM
|
||||
max_tokens = self.get_max_tokens()
|
||||
temperature = self.get_temperature()
|
||||
|
||||
response = await call_provider_with_retry(
|
||||
self.context, self.config_manager, prompt,
|
||||
max_tokens, temperature, umo
|
||||
)
|
||||
|
||||
if response is None:
|
||||
logger.error(f"{self.get_data_type()}分析调用LLM失败: provider返回None(重试失败)")
|
||||
return [], TokenUsage()
|
||||
|
||||
# 3. 提取token使用统计
|
||||
token_usage_dict = extract_token_usage(response)
|
||||
token_usage = TokenUsage(
|
||||
prompt_tokens=token_usage_dict["prompt_tokens"],
|
||||
completion_tokens=token_usage_dict["completion_tokens"],
|
||||
total_tokens=token_usage_dict["total_tokens"]
|
||||
)
|
||||
|
||||
# 4. 提取响应文本
|
||||
result_text = extract_response_text(response)
|
||||
logger.debug(f"{self.get_data_type()}分析原始响应: {result_text[:500]}...")
|
||||
|
||||
# 5. 尝试JSON解析
|
||||
success, parsed_data, error_msg = parse_json_response(result_text, self.get_data_type())
|
||||
|
||||
if success and parsed_data:
|
||||
# JSON解析成功,创建数据对象
|
||||
data_objects = self.create_data_objects(parsed_data)
|
||||
logger.info(f"{self.get_data_type()}分析成功,解析到 {len(data_objects)} 条数据")
|
||||
return data_objects, token_usage
|
||||
|
||||
# 6. JSON解析失败,使用正则表达式降级
|
||||
logger.warning(f"{self.get_data_type()}JSON解析失败,尝试正则表达式提取: {error_msg}")
|
||||
regex_data = self.extract_with_regex(result_text, self.get_max_count())
|
||||
|
||||
if regex_data:
|
||||
logger.info(f"{self.get_data_type()}正则表达式提取成功,获得 {len(regex_data)} 条数据")
|
||||
data_objects = self.create_data_objects(regex_data)
|
||||
return data_objects, token_usage
|
||||
else:
|
||||
# 最后的降级方案
|
||||
logger.warning(f"{self.get_data_type()}正则表达式提取失败,返回空列表")
|
||||
return [], token_usage
|
||||
|
||||
except Exception as e:
|
||||
logger.error(f"{self.get_data_type()}分析失败: {e}")
|
||||
return [], TokenUsage()
|
||||
|
||||
def get_max_tokens(self) -> int:
|
||||
"""
|
||||
获取最大token数,子类可重写
|
||||
|
||||
Returns:
|
||||
最大token数
|
||||
"""
|
||||
return 10000
|
||||
|
||||
def get_temperature(self) -> float:
|
||||
"""
|
||||
获取温度参数,子类可重写
|
||||
|
||||
Returns:
|
||||
温度参数
|
||||
"""
|
||||
return 0.6
|
||||
|
||||
|
||||
@@ -0,0 +1,195 @@
|
||||
"""
|
||||
金句分析模块
|
||||
专门处理群聊金句提取和分析
|
||||
"""
|
||||
|
||||
from typing import List, Dict, Tuple
|
||||
from datetime import datetime
|
||||
from astrbot.api import logger
|
||||
from ...models.data_models import GoldenQuote, TokenUsage
|
||||
from .base_analyzer import BaseAnalyzer
|
||||
from ..utils.json_utils import extract_golden_quotes_with_regex
|
||||
|
||||
|
||||
|
||||
class GoldenQuoteAnalyzer(BaseAnalyzer):
|
||||
"""
|
||||
金句分析器
|
||||
专门处理群聊金句的提取和分析
|
||||
"""
|
||||
|
||||
def get_data_type(self) -> str:
|
||||
"""获取数据类型标识"""
|
||||
return "金句"
|
||||
|
||||
def get_max_count(self) -> int:
|
||||
"""获取最大金句数量"""
|
||||
return self.config_manager.get_max_golden_quotes()
|
||||
|
||||
def get_max_tokens(self) -> int:
|
||||
"""获取最大token数"""
|
||||
return 1500
|
||||
|
||||
def get_temperature(self) -> float:
|
||||
"""获取温度参数"""
|
||||
return 0.7
|
||||
|
||||
def build_prompt(self, messages: List[Dict]) -> str:
|
||||
"""
|
||||
构建金句分析提示词
|
||||
|
||||
Args:
|
||||
messages: 圣经的文本消息列表
|
||||
|
||||
Returns:
|
||||
提示词字符串
|
||||
"""
|
||||
if not messages:
|
||||
return ""
|
||||
|
||||
# 构建消息文本
|
||||
messages_text = "\n".join([
|
||||
f"[{msg['time']}] {msg['sender']}: {msg['content']}"
|
||||
for msg in messages
|
||||
])
|
||||
|
||||
max_golden_quotes = self.get_max_count()
|
||||
|
||||
prompt = f"""
|
||||
请从以下群聊记录中挑选出{max_golden_quotes}句最具冲击力、最令人惊叹的"金句"。这些金句需满足:
|
||||
- 核心标准:**逆天的神人发言**,即具备颠覆常识的脑洞、逻辑跳脱的表达或强烈反差感的原创内容
|
||||
- 典型特征:包含某些争议话题元素、夸张类比、反常规结论、一本正经的"胡说八道"或突破语境的清奇思路,并且具备一定的冲击力,让人印象深刻。
|
||||
|
||||
对于每个金句,请提供:
|
||||
1. 原文内容(完整保留发言细节)
|
||||
2. 发言人昵称
|
||||
3. 选择理由(具体说明其"逆天"之处,如逻辑颠覆点/脑洞角度/反差感/争议话题元素)
|
||||
|
||||
此外,我将对你进行严格约束:
|
||||
- 优先筛选 **逆天指数最高** 的内容:发情、性压抑话题 > 争议话题 > 元素级 > 颠覆认知级 > 逻辑跳脱级 > 趣味调侃级,剔除单纯玩梗或网络热词堆砌的普通发言
|
||||
- 重点标记包含极端类比、反常识论证或无厘头结论的内容,并且包含一定的争议话题元素。
|
||||
|
||||
群聊记录:
|
||||
{messages_text}
|
||||
|
||||
请以JSON格式返回,格式如下:
|
||||
[
|
||||
{{
|
||||
"content": "金句原文",
|
||||
"sender": "发言人昵称",
|
||||
"reason": "选择这句话的理由(需明确说明逆天特质)"
|
||||
}}
|
||||
]
|
||||
"""
|
||||
return prompt
|
||||
|
||||
def extract_with_regex(self, result_text: str, max_count: int) -> List[Dict]:
|
||||
"""
|
||||
使用正则表达式提取金句信息
|
||||
|
||||
Args:
|
||||
result_text: LLM响应文本
|
||||
max_count: 最大提取数量
|
||||
|
||||
Returns:
|
||||
金句数据列表
|
||||
"""
|
||||
return extract_golden_quotes_with_regex(result_text, max_count)
|
||||
|
||||
def create_data_objects(self, quotes_data: List[Dict]) -> List[GoldenQuote]:
|
||||
"""
|
||||
创建金句对象列表
|
||||
|
||||
Args:
|
||||
quotes_data: 原始金句数据列表
|
||||
|
||||
Returns:
|
||||
GoldenQuote对象列表
|
||||
"""
|
||||
try:
|
||||
quotes = []
|
||||
max_quotes = self.get_max_count()
|
||||
|
||||
for quote_data in quotes_data[:max_quotes]:
|
||||
# 确保数据格式正确
|
||||
content = quote_data.get("content", "").strip()
|
||||
sender = quote_data.get("sender", "").strip()
|
||||
reason = quote_data.get("reason", "").strip()
|
||||
|
||||
# 验证必要字段
|
||||
if not content or not sender or not reason:
|
||||
logger.warning(f"金句数据格式不完整,跳过: {quote_data}")
|
||||
continue
|
||||
|
||||
quotes.append(GoldenQuote(
|
||||
content=content,
|
||||
sender=sender,
|
||||
reason=reason
|
||||
))
|
||||
|
||||
return quotes
|
||||
|
||||
except Exception as e:
|
||||
logger.error(f"创建金句对象失败: {e}")
|
||||
return []
|
||||
|
||||
def extract_interesting_messages(self, messages: List[Dict]) -> List[Dict]:
|
||||
"""
|
||||
提取圣经的文本消息
|
||||
|
||||
Args:
|
||||
messages: 群聊消息列表
|
||||
|
||||
Returns:
|
||||
圣经的文本消息列表
|
||||
"""
|
||||
try:
|
||||
interesting_messages = []
|
||||
|
||||
for msg in messages:
|
||||
sender = msg.get("sender", {})
|
||||
nickname = sender.get("nickname", "") or sender.get("card", "")
|
||||
msg_time = datetime.fromtimestamp(msg.get("time", 0)).strftime("%H:%M")
|
||||
|
||||
for content in msg.get("message", []):
|
||||
if content.get("type") == "text":
|
||||
text = content.get("data", {}).get("text", "").strip()
|
||||
# 过滤长度适中、可能圣经的消息
|
||||
if 5 <= len(text) <= 100 and not text.startswith(("http", "www", "/")):
|
||||
interesting_messages.append({
|
||||
"sender": nickname,
|
||||
"time": msg_time,
|
||||
"content": text
|
||||
})
|
||||
|
||||
return interesting_messages
|
||||
|
||||
except Exception as e:
|
||||
logger.error(f"提取圣经消息失败: {e}")
|
||||
return []
|
||||
|
||||
async def analyze_golden_quotes(self, messages: List[Dict], umo: str = None) -> Tuple[List[GoldenQuote], TokenUsage]:
|
||||
"""
|
||||
分析群聊金句
|
||||
|
||||
Args:
|
||||
messages: 群聊消息列表
|
||||
umo: 模型唯一标识符
|
||||
|
||||
Returns:
|
||||
(金句列表, Token使用统计)
|
||||
"""
|
||||
try:
|
||||
# 提取圣经的文本消息
|
||||
interesting_messages = self.extract_interesting_messages(messages)
|
||||
|
||||
if not interesting_messages:
|
||||
logger.info("没有符合条件的圣经消息,返回空结果")
|
||||
return [], TokenUsage()
|
||||
|
||||
logger.info(f"开始从 {len(interesting_messages)} 条圣经消息中提取金句")
|
||||
return await self.analyze(interesting_messages, umo)
|
||||
|
||||
except Exception as e:
|
||||
logger.error(f"金句分析失败: {e}")
|
||||
return [], TokenUsage()
|
||||
@@ -0,0 +1,207 @@
|
||||
"""
|
||||
话题分析模块
|
||||
专门处理群聊话题分析
|
||||
"""
|
||||
|
||||
from typing import List, Dict, Tuple
|
||||
from datetime import datetime
|
||||
import re
|
||||
from astrbot.api import logger
|
||||
from ...models.data_models import SummaryTopic, TokenUsage
|
||||
from .base_analyzer import BaseAnalyzer
|
||||
from ..utils.json_utils import extract_topics_with_regex
|
||||
|
||||
|
||||
class TopicAnalyzer(BaseAnalyzer):
|
||||
"""
|
||||
话题分析器
|
||||
专门处理群聊话题的提取和分析
|
||||
"""
|
||||
|
||||
def get_data_type(self) -> str:
|
||||
"""获取数据类型标识"""
|
||||
return "话题"
|
||||
|
||||
def get_max_count(self) -> int:
|
||||
"""获取最大话题数量"""
|
||||
return self.config_manager.get_max_topics()
|
||||
|
||||
def get_max_tokens(self) -> int:
|
||||
"""获取最大token数"""
|
||||
return 10000
|
||||
|
||||
def get_temperature(self) -> float:
|
||||
"""获取温度参数"""
|
||||
return 0.6
|
||||
|
||||
def build_prompt(self, messages: List[Dict]) -> str:
|
||||
"""
|
||||
构建话题分析提示词
|
||||
|
||||
Args:
|
||||
messages: 群聊消息列表
|
||||
|
||||
Returns:
|
||||
提示词字符串
|
||||
"""
|
||||
# 提取文本消息
|
||||
text_messages = []
|
||||
for msg in messages:
|
||||
sender = msg.get("sender", {})
|
||||
nickname = sender.get("nickname", "") or sender.get("card", "")
|
||||
msg_time = datetime.fromtimestamp(msg.get("time", 0)).strftime("%H:%M")
|
||||
|
||||
for content in msg.get("message", []):
|
||||
if content.get("type") == "text":
|
||||
text = content.get("data", {}).get("text", "").strip()
|
||||
if text and len(text) > 2 and not text.startswith("/"):
|
||||
# 清理消息内容
|
||||
text = text.replace('"', '"').replace('"', '"')
|
||||
text = text.replace(''', "'").replace(''', "'")
|
||||
text = text.replace('\n', ' ').replace('\r', ' ')
|
||||
text = text.replace('\t', ' ')
|
||||
text = re.sub(r'[\x00-\x1f\x7f-\x9f]', '', text)
|
||||
text_messages.append({
|
||||
"sender": nickname,
|
||||
"time": msg_time,
|
||||
"content": text.strip()
|
||||
})
|
||||
|
||||
if not text_messages:
|
||||
return ""
|
||||
|
||||
# 构建消息文本
|
||||
messages_text = "\n".join([
|
||||
f"[{msg['time']}] {msg['sender']}: {msg['content']}"
|
||||
for msg in text_messages
|
||||
])
|
||||
|
||||
max_topics = self.get_max_count()
|
||||
|
||||
prompt = f"""
|
||||
你是一个帮我进行群聊信息总结的助手,生成总结内容时,你需要严格遵守下面的几个准则:
|
||||
请分析接下来提供的群聊记录,提取出最多{max_topics}个主要话题。
|
||||
|
||||
对于每个话题,请提供:
|
||||
1. 话题名称(突出主题内容,尽量简明扼要)
|
||||
2. 主要参与者(最多5人)
|
||||
3. 话题详细描述(包含关键信息和结论)
|
||||
|
||||
注意:
|
||||
- 对于比较有价值的点,稍微用一两句话详细讲讲,比如不要生成 "Nolan 和 SOV 讨论了 galgame 中关于性符号的衍生情况" 这种宽泛的内容,而是生成更加具体的讨论内容,让其他人只看这个消息就能知道讨论中有价值的,有营养的信息。
|
||||
- 对于其中的部分信息,你需要特意提到主题施加的主体是谁,是哪个群友做了什么事情,而不要直接生成和群友没有关系的语句。
|
||||
- 对于每一条总结,尽量讲清楚前因后果,以及话题的结论,是什么,为什么,怎么做,如果用户没有讲到细节,则可以不用这么做。
|
||||
|
||||
群聊记录:
|
||||
{messages_text}
|
||||
|
||||
重要:必须返回标准JSON格式,严格遵守以下规则:
|
||||
1. 只使用英文双引号 " 不要使用中文引号 " "
|
||||
2. 字符串内容中的引号必须转义为 \"
|
||||
3. 多个对象之间用逗号分隔
|
||||
4. 数组元素之间用逗号分隔
|
||||
5. 不要在JSON外添加任何文字说明
|
||||
6. 描述内容避免使用特殊符号,用普通文字表达
|
||||
|
||||
请严格按照以下JSON格式返回,确保可以被标准JSON解析器解析:
|
||||
[
|
||||
{{
|
||||
"topic": "话题名称",
|
||||
"contributors": ["用户1", "用户2"],
|
||||
"detail": "话题描述内容"
|
||||
}},
|
||||
{{
|
||||
"topic": "另一个话题",
|
||||
"contributors": ["用户3", "用户4"],
|
||||
"detail": "另一个话题的描述"
|
||||
}}
|
||||
]
|
||||
|
||||
注意:返回的内容必须是纯JSON,不要包含markdown代码块标记或其他格式
|
||||
"""
|
||||
return prompt
|
||||
|
||||
def extract_with_regex(self, result_text: str, max_topics: int) -> List[Dict]:
|
||||
"""
|
||||
使用正则表达式提取话题信息
|
||||
|
||||
Args:
|
||||
result_text: LLM响应文本
|
||||
max_topics: 最大话题数量
|
||||
|
||||
Returns:
|
||||
话题数据列表
|
||||
"""
|
||||
return extract_topics_with_regex(result_text, max_topics)
|
||||
|
||||
def create_data_objects(self, topics_data: List[Dict]) -> List[SummaryTopic]:
|
||||
"""
|
||||
创建话题对象列表
|
||||
|
||||
Args:
|
||||
topics_data: 原始话题数据列表
|
||||
|
||||
Returns:
|
||||
SummaryTopic对象列表
|
||||
"""
|
||||
try:
|
||||
topics = []
|
||||
max_topics = self.get_max_count()
|
||||
|
||||
for topic_data in topics_data[:max_topics]:
|
||||
# 确保数据格式正确
|
||||
topic_name = topic_data.get("topic", "").strip()
|
||||
contributors = topic_data.get("contributors", [])
|
||||
detail = topic_data.get("detail", "").strip()
|
||||
|
||||
# 验证必要字段
|
||||
if not topic_name or not detail:
|
||||
logger.warning(f"话题数据格式不完整,跳过: {topic_data}")
|
||||
continue
|
||||
|
||||
# 确保参与者列表有效
|
||||
if not contributors or not isinstance(contributors, list):
|
||||
contributors = ["群友"]
|
||||
else:
|
||||
# 清理参与者名称
|
||||
contributors = [str(c).strip() for c in contributors if c and str(c).strip()]
|
||||
if not contributors:
|
||||
contributors = ["群友"]
|
||||
|
||||
topics.append(SummaryTopic(
|
||||
topic=topic_name,
|
||||
contributors=contributors[:5], # 最多5个参与者
|
||||
detail=detail
|
||||
))
|
||||
|
||||
return topics
|
||||
|
||||
except Exception as e:
|
||||
logger.error(f"创建话题对象失败: {e}")
|
||||
return []
|
||||
|
||||
async def analyze_topics(self, messages: List[Dict], umo: str = None) -> Tuple[List[SummaryTopic], TokenUsage]:
|
||||
"""
|
||||
分析群聊话题
|
||||
|
||||
Args:
|
||||
messages: 群聊消息列表
|
||||
umo: 模型唯一标识符
|
||||
|
||||
Returns:
|
||||
(话题列表, Token使用统计)
|
||||
"""
|
||||
try:
|
||||
# 提取文本消息
|
||||
text_messages = self.extract_text_messages(messages)
|
||||
|
||||
if not text_messages:
|
||||
logger.info("没有有效的文本消息,返回空结果")
|
||||
return [], TokenUsage()
|
||||
|
||||
logger.info(f"开始分析 {len(text_messages)} 条文本消息中的话题")
|
||||
return await self.analyze(text_messages, umo)
|
||||
|
||||
except Exception as e:
|
||||
logger.error(f"话题分析失败: {e}")
|
||||
return [], TokenUsage()
|
||||
@@ -0,0 +1,222 @@
|
||||
"""
|
||||
用户称号分析模块
|
||||
专门处理用户称号和MBTI类型分析
|
||||
"""
|
||||
|
||||
from typing import List, Dict, Tuple
|
||||
from astrbot.api import logger
|
||||
from ...models.data_models import UserTitle, TokenUsage
|
||||
from .base_analyzer import BaseAnalyzer
|
||||
from ..utils.json_utils import extract_user_titles_with_regex
|
||||
|
||||
|
||||
class UserTitleAnalyzer(BaseAnalyzer):
|
||||
"""
|
||||
用户称号分析器
|
||||
专门处理用户称号分配和MBTI类型分析
|
||||
"""
|
||||
|
||||
def get_data_type(self) -> str:
|
||||
"""获取数据类型标识"""
|
||||
return "用户称号"
|
||||
|
||||
def get_max_count(self) -> int:
|
||||
"""获取最大用户称号数量"""
|
||||
return self.config_manager.get_max_user_titles()
|
||||
|
||||
def get_max_tokens(self) -> int:
|
||||
"""获取最大token数"""
|
||||
return 1500
|
||||
|
||||
def get_temperature(self) -> float:
|
||||
"""获取温度参数"""
|
||||
return 0.5
|
||||
|
||||
def build_prompt(self, user_data: Dict) -> str:
|
||||
"""
|
||||
构建用户称号分析提示词
|
||||
|
||||
Args:
|
||||
user_data: 用户数据字典,包含用户统计信息
|
||||
|
||||
Returns:
|
||||
提示词字符串
|
||||
"""
|
||||
user_summaries = user_data.get("user_summaries", [])
|
||||
|
||||
if not user_summaries:
|
||||
return ""
|
||||
|
||||
# 构建用户数据文本
|
||||
users_text = "\n".join([
|
||||
f"- {user['name']} (QQ:{user['qq']}): "
|
||||
f"发言{user['message_count']}条, 平均{user['avg_chars']}字, "
|
||||
f"表情比例{user['emoji_ratio']}, 夜间发言比例{user['night_ratio']}, "
|
||||
f"回复比例{user['reply_ratio']}"
|
||||
for user in user_summaries
|
||||
])
|
||||
|
||||
prompt = f"""
|
||||
请为以下群友分配合适的称号和MBTI类型。每个人只能有一个称号,每个称号只能给一个人。
|
||||
|
||||
可选称号:
|
||||
- 龙王: 发言频繁但内容轻松的人
|
||||
- 技术专家: 经常讨论技术话题的人
|
||||
- 夜猫子: 经常在深夜发言的人
|
||||
- 表情包军火库: 经常发表情的人
|
||||
- 沉默终结者: 经常开启话题的人
|
||||
- 评论家: 平均发言长度很长的人
|
||||
- 阳角: 在群里很有影响力的人
|
||||
- 互动达人: 经常回复别人的人
|
||||
- ... (你可以自行进行拓展添加)
|
||||
|
||||
用户数据:
|
||||
{users_text}
|
||||
|
||||
请以JSON格式返回,格式如下:
|
||||
[
|
||||
{{
|
||||
"name": "用户名",
|
||||
"qq": 123456789,
|
||||
"title": "称号",
|
||||
"mbti": "MBTI类型",
|
||||
"reason": "获得此称号的原因"
|
||||
}}
|
||||
]
|
||||
"""
|
||||
return prompt
|
||||
|
||||
def extract_with_regex(self, result_text: str, max_count: int) -> List[Dict]:
|
||||
"""
|
||||
使用正则表达式提取用户称号信息
|
||||
|
||||
Args:
|
||||
result_text: LLM响应文本
|
||||
max_count: 最大提取数量
|
||||
|
||||
Returns:
|
||||
用户称号数据列表
|
||||
"""
|
||||
return extract_user_titles_with_regex(result_text, max_count)
|
||||
|
||||
def create_data_objects(self, titles_data: List[Dict]) -> List[UserTitle]:
|
||||
"""
|
||||
创建用户称号对象列表
|
||||
|
||||
Args:
|
||||
titles_data: 原始用户称号数据列表
|
||||
|
||||
Returns:
|
||||
UserTitle对象列表
|
||||
"""
|
||||
try:
|
||||
titles = []
|
||||
max_titles = self.get_max_count()
|
||||
|
||||
for title_data in titles_data[:max_titles]:
|
||||
# 确保数据格式正确
|
||||
name = title_data.get("name", "").strip()
|
||||
qq = title_data.get("qq")
|
||||
title = title_data.get("title", "").strip()
|
||||
mbti = title_data.get("mbti", "").strip()
|
||||
reason = title_data.get("reason", "").strip()
|
||||
|
||||
# 验证必要字段
|
||||
if not name or not title or not mbti or not reason:
|
||||
logger.warning(f"用户称号数据格式不完整,跳过: {title_data}")
|
||||
continue
|
||||
|
||||
# 验证QQ号格式
|
||||
try:
|
||||
qq = int(qq)
|
||||
except (ValueError, TypeError):
|
||||
logger.warning(f"QQ号格式无效,跳过: {qq}")
|
||||
continue
|
||||
|
||||
titles.append(UserTitle(
|
||||
name=name,
|
||||
qq=qq,
|
||||
title=title,
|
||||
mbti=mbti,
|
||||
reason=reason
|
||||
))
|
||||
|
||||
return titles
|
||||
|
||||
except Exception as e:
|
||||
logger.error(f"创建用户称号对象失败: {e}")
|
||||
return []
|
||||
|
||||
def prepare_user_data(self, messages: List[Dict], user_analysis: Dict) -> Dict:
|
||||
"""
|
||||
准备用户数据
|
||||
|
||||
Args:
|
||||
messages: 群聊消息列表
|
||||
user_analysis: 用户分析统计
|
||||
|
||||
Returns:
|
||||
准备好的用户数据字典
|
||||
"""
|
||||
try:
|
||||
user_summaries = []
|
||||
|
||||
for user_id, stats in user_analysis.items():
|
||||
if stats["message_count"] < 5: # 过滤活跃度太低的用户
|
||||
continue
|
||||
|
||||
# 分析用户特征
|
||||
night_messages = sum(stats["hours"][h] for h in range(0, 6))
|
||||
day_messages = stats["message_count"] - night_messages
|
||||
avg_chars = stats["char_count"] / stats["message_count"] if stats["message_count"] > 0 else 0
|
||||
|
||||
user_summaries.append({
|
||||
"name": stats["nickname"],
|
||||
"qq": int(user_id),
|
||||
"message_count": stats["message_count"],
|
||||
"avg_chars": round(avg_chars, 1),
|
||||
"emoji_ratio": round(stats["emoji_count"] / stats["message_count"], 2),
|
||||
"night_ratio": round(night_messages / stats["message_count"], 2),
|
||||
"reply_ratio": round(stats["reply_count"] / stats["message_count"], 2)
|
||||
})
|
||||
|
||||
if not user_summaries:
|
||||
return {"user_summaries": []}
|
||||
|
||||
# 按消息数量排序,取前N名
|
||||
max_user_titles = self.get_max_count()
|
||||
user_summaries.sort(key=lambda x: x["message_count"], reverse=True)
|
||||
user_summaries = user_summaries[:max_user_titles]
|
||||
|
||||
return {"user_summaries": user_summaries}
|
||||
|
||||
except Exception as e:
|
||||
logger.error(f"准备用户数据失败: {e}")
|
||||
return {"user_summaries": []}
|
||||
|
||||
async def analyze_user_titles(self, messages: List[Dict], user_analysis: Dict, umo: str = None) -> Tuple[List[UserTitle], TokenUsage]:
|
||||
"""
|
||||
分析用户称号
|
||||
|
||||
Args:
|
||||
messages: 群聊消息列表
|
||||
user_analysis: 用户分析统计
|
||||
umo: 模型唯一标识符
|
||||
|
||||
Returns:
|
||||
(用户称号列表, Token使用统计)
|
||||
"""
|
||||
try:
|
||||
# 准备用户数据
|
||||
user_data = self.prepare_user_data(messages, user_analysis)
|
||||
|
||||
if not user_data["user_summaries"]:
|
||||
logger.info("没有符合条件的用户,返回空结果")
|
||||
return [], TokenUsage()
|
||||
|
||||
logger.info(f"开始分析 {len(user_data['user_summaries'])} 个用户的称号")
|
||||
return await self.analyze(user_data, umo)
|
||||
|
||||
except Exception as e:
|
||||
logger.error(f"用户称号分析失败: {e}")
|
||||
return [], TokenUsage()
|
||||
+119
-554
@@ -1,580 +1,145 @@
|
||||
"""
|
||||
LLM分析器模块
|
||||
负责使用LLM进行话题分析、用户称号分析和金句分析
|
||||
负责协调各个分析器进行话题分析、用户称号分析和金句分析
|
||||
"""
|
||||
|
||||
import json
|
||||
import re
|
||||
from datetime import datetime
|
||||
import asyncio
|
||||
from typing import List, Dict, Tuple
|
||||
from astrbot.api import logger
|
||||
from ...src.models.data_models import SummaryTopic, UserTitle, GoldenQuote, TokenUsage
|
||||
from ..models.data_models import SummaryTopic, UserTitle, GoldenQuote, TokenUsage
|
||||
from .analyzers.topic_analyzer import TopicAnalyzer
|
||||
from .analyzers.user_title_analyzer import UserTitleAnalyzer
|
||||
from .analyzers.golden_quote_analyzer import GoldenQuoteAnalyzer
|
||||
from .utils.llm_utils import call_provider_with_retry
|
||||
from .utils.json_utils import fix_json
|
||||
from .utils.json_utils import extract_topics_with_regex, extract_user_titles_with_regex, extract_golden_quotes_with_regex
|
||||
|
||||
|
||||
class LLMAnalyzer:
|
||||
"""LLM分析器"""
|
||||
|
||||
"""
|
||||
LLM分析器
|
||||
作为统一入口,协调各个专门的分析器进行不同类型的分析
|
||||
保持向后兼容性,提供原有的接口
|
||||
"""
|
||||
|
||||
def __init__(self, context, config_manager):
|
||||
"""
|
||||
初始化LLM分析器
|
||||
|
||||
Args:
|
||||
context: AstrBot上下文对象
|
||||
config_manager: 配置管理器
|
||||
"""
|
||||
self.context = context
|
||||
self.config_manager = config_manager
|
||||
|
||||
async def _call_provider_with_retry(self, provider, prompt: str, max_tokens: int, temperature: float, umo: str = None):
|
||||
"""
|
||||
调用LLM提供者,带超时、重试与退避。支持自定义服务商。
|
||||
|
||||
Args:
|
||||
provider: LLM服务商实例或None。
|
||||
prompt (str): 输入的提示语。
|
||||
max_tokens (int): 最大生成token数。
|
||||
temperature (float): 采样温度。
|
||||
umo (str, optional): 指定使用的模型唯一标识符(Unique Model Object),
|
||||
用于选择特定的LLM服务商或模型。格式通常为字符串,例如 "gpt-3.5-turbo"。
|
||||
如果为None,则使用默认模型。
|
||||
|
||||
Returns:
|
||||
LLM生成的结果。
|
||||
"""
|
||||
timeout = self.config_manager.get_llm_timeout()
|
||||
retries = self.config_manager.get_llm_retries()
|
||||
backoff = self.config_manager.get_llm_backoff()
|
||||
|
||||
# 获取自定义服务商参数
|
||||
custom_api_key = self.config_manager.get_custom_api_key()
|
||||
custom_api_base = self.config_manager.get_custom_api_base_url()
|
||||
custom_model = self.config_manager.get_custom_model_name()
|
||||
|
||||
last_exc = None
|
||||
for attempt in range(1, retries + 1):
|
||||
try:
|
||||
if custom_api_key and custom_api_base and custom_model:
|
||||
logger.info(f"使用自定义LLM提供商: {custom_api_base} model={custom_model}")
|
||||
import aiohttp
|
||||
async with aiohttp.ClientSession() as session:
|
||||
headers = {
|
||||
"Authorization": f"Bearer {custom_api_key}",
|
||||
"Content-Type": "application/json"
|
||||
}
|
||||
payload = {
|
||||
"model": custom_model,
|
||||
"messages": [{"role": "user", "content": prompt}],
|
||||
"max_tokens": max_tokens,
|
||||
"temperature": temperature
|
||||
}
|
||||
aio_timeout = aiohttp.ClientTimeout(total=timeout)
|
||||
async with session.post(custom_api_base, json=payload, headers=headers, timeout=aio_timeout) as resp:
|
||||
if resp.status != 200:
|
||||
error_text = await resp.text()
|
||||
logger.error(f"自定义LLM服务商请求失败: HTTP {resp.status}, 内容: {error_text}")
|
||||
try:
|
||||
response_json = await resp.json()
|
||||
except Exception as json_err:
|
||||
error_text = await resp.text()
|
||||
logger.error(f"自定义LLM服务商响应JSON解析失败: {json_err}, 内容: {error_text}")
|
||||
return None
|
||||
# 兼容 OpenAI 格式,安全访问嵌套字段
|
||||
content = None
|
||||
try:
|
||||
choices = response_json.get("choices")
|
||||
if choices and isinstance(choices, list) and len(choices) > 0:
|
||||
message = choices[0].get("message")
|
||||
if message and isinstance(message, dict):
|
||||
content = message.get("content")
|
||||
if content is None:
|
||||
logger.error(f"自定义LLM响应格式异常: {response_json}")
|
||||
return None
|
||||
except Exception as key_err:
|
||||
logger.error(f"自定义LLM响应结构解析失败: {key_err}, 响应内容: {response_json}")
|
||||
return None
|
||||
# 构造一个兼容原有逻辑的对象
|
||||
class CustomResponse:
|
||||
completion_text = content
|
||||
raw_completion = response_json
|
||||
return CustomResponse()
|
||||
else:
|
||||
# 确保使用当前指定的模型
|
||||
if provider is None:
|
||||
provider = self.context.get_using_provider(umo=umo)
|
||||
provider_id = 'unknown'
|
||||
if provider:
|
||||
try:
|
||||
meta = provider.meta()
|
||||
provider_id = meta.id
|
||||
except Exception as e:
|
||||
logger.debug(f"获取提供商ID失败: {e}")
|
||||
logger.info(f"获取到的 provider ID: {provider_id}")
|
||||
if not provider or provider_id == 'unknown':
|
||||
logger.warning(f"获取的提供商不正确 (Provider ID: {provider_id})")
|
||||
|
||||
|
||||
logger.info(f"使用LLM provider: {provider}")
|
||||
if not provider:
|
||||
logger.error("provider 为空,无法调用 text_chat,直接返回 None")
|
||||
return None
|
||||
coro = provider.text_chat(prompt=prompt, max_tokens=max_tokens, temperature=temperature)
|
||||
return await asyncio.wait_for(coro, timeout=timeout)
|
||||
except asyncio.TimeoutError as e:
|
||||
last_exc = e
|
||||
logger.warning(f"LLM请求超时: 第{attempt}次, timeout={timeout}s")
|
||||
except Exception as e:
|
||||
last_exc = e
|
||||
logger.warning(f"LLM请求失败: 第{attempt}次, 错误: {last_exc}")
|
||||
# 若非最后一次,等待退避后重试
|
||||
if attempt < retries:
|
||||
await asyncio.sleep(backoff * attempt)
|
||||
|
||||
# 最终仍失败,记录错误并返回 None 由调用方处理降级,避免抛出异常
|
||||
logger.error(f"LLM请求全部重试失败: {last_exc}")
|
||||
return None
|
||||
|
||||
|
||||
# 初始化各个专门的分析器
|
||||
self.topic_analyzer = TopicAnalyzer(context, config_manager)
|
||||
self.user_title_analyzer = UserTitleAnalyzer(context, config_manager)
|
||||
self.golden_quote_analyzer = GoldenQuoteAnalyzer(context, config_manager)
|
||||
|
||||
async def analyze_topics(self, messages: List[Dict], umo: str = None) -> Tuple[List[SummaryTopic], TokenUsage]:
|
||||
"""使用LLM分析话题"""
|
||||
"""
|
||||
使用LLM分析话题
|
||||
保持原有接口,委托给专门的TopicAnalyzer处理
|
||||
|
||||
Args:
|
||||
messages: 群聊消息列表
|
||||
umo: 模型唯一标识符
|
||||
|
||||
Returns:
|
||||
(话题列表, Token使用统计)
|
||||
"""
|
||||
try:
|
||||
# 提取文本消息
|
||||
text_messages = []
|
||||
for msg in messages:
|
||||
sender = msg.get("sender", {})
|
||||
nickname = sender.get("nickname", "") or sender.get("card", "")
|
||||
msg_time = datetime.fromtimestamp(msg.get("time", 0)).strftime("%H:%M")
|
||||
|
||||
for content in msg.get("message", []):
|
||||
if content.get("type") == "text":
|
||||
text = content.get("data", {}).get("text", "").strip()
|
||||
if text and len(text) > 2 and not text.startswith(("/")):
|
||||
text_messages.append({
|
||||
"sender": nickname,
|
||||
"time": msg_time,
|
||||
"content": text
|
||||
})
|
||||
|
||||
if not text_messages:
|
||||
return [], TokenUsage()
|
||||
|
||||
# 构建LLM提示词,清理消息内容
|
||||
def clean_message_content(content):
|
||||
"""清理消息内容,移除可能影响JSON解析的字符"""
|
||||
# 替换中文引号
|
||||
content = content.replace('"', '"').replace('"', '"')
|
||||
content = content.replace(''', "'").replace(''', "'")
|
||||
# 移除或替换其他特殊字符
|
||||
content = content.replace('\n', ' ').replace('\r', ' ')
|
||||
content = content.replace('\t', ' ')
|
||||
# 移除可能的控制字符
|
||||
content = re.sub(r'[\x00-\x1f\x7f-\x9f]', '', content)
|
||||
return content.strip()
|
||||
|
||||
messages_text = "\n".join([
|
||||
f"[{msg['time']}] {msg['sender']}: {clean_message_content(msg['content'])}"
|
||||
for msg in text_messages
|
||||
])
|
||||
|
||||
max_topics = self.config_manager.get_max_topics()
|
||||
prompt = f"""
|
||||
你是一个帮我进行群聊信息总结的助手,生成总结内容时,你需要严格遵守下面的几个准则:
|
||||
请分析接下来提供的群聊记录,提取出最多{max_topics}个主要话题。
|
||||
|
||||
对于每个话题,请提供:
|
||||
1. 话题名称(突出主题内容,尽量简明扼要)
|
||||
2. 主要参与者(最多5人)
|
||||
3. 话题详细描述(包含关键信息和结论)
|
||||
|
||||
注意:
|
||||
- 对于比较有价值的点,稍微用一两句话详细讲讲,比如不要生成 "Nolan 和 SOV 讨论了 galgame 中关于性符号的衍生情况" 这种宽泛的内容,而是生成更加具体的讨论内容,让其他人只看这个消息就能知道讨论中有价值的,有营养的信息。
|
||||
- 对于其中的部分信息,你需要特意提到主题施加的主体是谁,是哪个群友做了什么事情,而不要直接生成和群友没有关系的语句。
|
||||
- 对于每一条总结,尽量讲清楚前因后果,以及话题的结论,是什么,为什么,怎么做,如果用户没有讲到细节,则可以不用这么做。
|
||||
|
||||
群聊记录:
|
||||
{messages_text}
|
||||
|
||||
重要:必须返回标准JSON格式,严格遵守以下规则:
|
||||
1. 只使用英文双引号 " 不要使用中文引号 " "
|
||||
2. 字符串内容中的引号必须转义为 \"
|
||||
3. 多个对象之间用逗号分隔
|
||||
4. 数组元素之间用逗号分隔
|
||||
5. 不要在JSON外添加任何文字说明
|
||||
6. 描述内容避免使用特殊符号,用普通文字表达
|
||||
|
||||
请严格按照以下JSON格式返回,确保可以被标准JSON解析器解析:
|
||||
[
|
||||
{{
|
||||
"topic": "话题名称",
|
||||
"contributors": ["用户1", "用户2"],
|
||||
"detail": "话题描述内容"
|
||||
}},
|
||||
{{
|
||||
"topic": "另一个话题",
|
||||
"contributors": ["用户3", "用户4"],
|
||||
"detail": "另一个话题的描述"
|
||||
}}
|
||||
]
|
||||
|
||||
注意:返回的内容必须是纯JSON,不要包含markdown代码块标记或其他格式
|
||||
"""
|
||||
|
||||
# 调用LLM
|
||||
response = await self._call_provider_with_retry(None, prompt, max_tokens=10000, temperature=0.6, umo=umo)
|
||||
if response is None:
|
||||
logger.error("话题分析调用LLM失败: provider返回None(重试失败)")
|
||||
return [], TokenUsage()
|
||||
|
||||
# 提取token使用统计
|
||||
token_usage = TokenUsage()
|
||||
# 安全地提取 usage,避免 response.raw_completion.usage 为 None 导致的 AttributeError
|
||||
usage = None
|
||||
if getattr(response, 'raw_completion', None) is not None:
|
||||
usage = getattr(response.raw_completion, 'usage', None)
|
||||
if usage:
|
||||
token_usage.prompt_tokens = getattr(usage, 'prompt_tokens', 0) or 0
|
||||
token_usage.completion_tokens = getattr(usage, 'completion_tokens', 0) or 0
|
||||
token_usage.total_tokens = getattr(usage, 'total_tokens', 0) or 0
|
||||
|
||||
# 解析响应
|
||||
if hasattr(response, 'completion_text'):
|
||||
result_text = response.completion_text
|
||||
else:
|
||||
result_text = str(response)
|
||||
|
||||
# 尝试解析JSON
|
||||
try:
|
||||
# 提取JSON部分
|
||||
json_match = re.search(r'\[.*?\]', result_text, re.DOTALL)
|
||||
if json_match:
|
||||
json_text = json_match.group()
|
||||
logger.debug(f"话题分析JSON原文: {json_text[:500]}...")
|
||||
|
||||
# 强化JSON清理和修复
|
||||
json_text = self._fix_json(json_text)
|
||||
logger.debug(f"修复后的JSON: {json_text[:300]}...")
|
||||
|
||||
topics_data = json.loads(json_text)
|
||||
topics = [SummaryTopic(**topic) for topic in topics_data[:max_topics]]
|
||||
logger.info(f"话题分析成功,解析到 {len(topics)} 个话题")
|
||||
return topics, token_usage
|
||||
else:
|
||||
logger.warning(f"话题分析响应中未找到JSON格式,响应内容: {result_text[:200]}...")
|
||||
except json.JSONDecodeError as e:
|
||||
logger.warning(f"话题分析JSON解析失败: {e}")
|
||||
logger.debug(f"修复后的JSON: {json_text if 'json_text' in locals() else 'N/A'}")
|
||||
logger.debug(f"原始响应: {result_text}")
|
||||
|
||||
# 如果JSON解析失败,尝试用正则表达式提取话题信息
|
||||
topics = self._extract_topics_with_regex(result_text, max_topics)
|
||||
if topics:
|
||||
logger.info(f"正则表达式提取成功,获得 {len(topics)} 个话题,话题分析 warning 可忽略")
|
||||
return topics, token_usage
|
||||
else:
|
||||
# 最后的降级方案
|
||||
logger.info("正则表达式提取失败,使用默认话题...")
|
||||
return [SummaryTopic(
|
||||
topic="群聊讨论",
|
||||
contributors=["群友"],
|
||||
detail="今日群聊内容丰富,涵盖多个话题"
|
||||
)], token_usage
|
||||
|
||||
return [], token_usage
|
||||
|
||||
logger.info("开始话题分析")
|
||||
return await self.topic_analyzer.analyze_topics(messages, umo)
|
||||
except Exception as e:
|
||||
logger.error(f"话题分析失败: {e}")
|
||||
return [], TokenUsage()
|
||||
|
||||
def _fix_json(self, text: str) -> str:
|
||||
"""修复JSON格式问题"""
|
||||
# 移除markdown代码块标记
|
||||
text = re.sub(r'```json\s*', '', text)
|
||||
text = re.sub(r'```\s*$', '', text)
|
||||
|
||||
# 基础清理
|
||||
text = text.replace('\n', ' ').replace('\r', ' ')
|
||||
text = re.sub(r'\s+', ' ', text)
|
||||
|
||||
# 替换中文引号为英文引号
|
||||
text = text.replace('"', '"').replace('"', '"')
|
||||
text = text.replace(''', "'").replace(''', "'")
|
||||
|
||||
# 处理字符串内容中的特殊字符
|
||||
# 转义字符串内的双引号
|
||||
def escape_quotes_in_strings(match):
|
||||
content = match.group(1)
|
||||
# 转义内部的双引号
|
||||
content = content.replace('"', '\\"')
|
||||
return f'"{content}"'
|
||||
|
||||
# 先处理字段值中的引号
|
||||
text = re.sub(r'"([^"]*(?:"[^"]*)*)"', escape_quotes_in_strings, text)
|
||||
|
||||
# 修复截断的JSON
|
||||
if not text.endswith(']'):
|
||||
last_complete = text.rfind('}')
|
||||
if last_complete > 0:
|
||||
text = text[:last_complete + 1] + ']'
|
||||
|
||||
# 修复常见的JSON格式问题
|
||||
# 1. 修复缺失的逗号
|
||||
text = re.sub(r'}\s*{', '}, {', text)
|
||||
|
||||
# 2. 确保字段名有引号
|
||||
text = re.sub(r'([{,]\s*)([a-zA-Z_][a-zA-Z0-9_]*)\s*:', r'\1"\2":', text)
|
||||
|
||||
# 3. 移除多余的逗号
|
||||
text = re.sub(r',\s*}', '}', text)
|
||||
text = re.sub(r',\s*]', ']', text)
|
||||
|
||||
return text
|
||||
|
||||
def _extract_topics_with_regex(self, result_text: str, max_topics: int) -> List[SummaryTopic]:
|
||||
"""使用正则表达式提取话题信息"""
|
||||
try:
|
||||
topics = []
|
||||
|
||||
# 更强的正则表达式提取话题信息,处理转义字符
|
||||
# 匹配每个完整的话题对象
|
||||
topic_pattern = r'\{\s*"topic":\s*"([^"]+)"\s*,\s*"contributors":\s*\[([^\]]+)\]\s*,\s*"detail":\s*"([^"]*(?:\\.[^"]*)*)"\s*\}'
|
||||
matches = re.findall(topic_pattern, result_text, re.DOTALL)
|
||||
|
||||
if not matches:
|
||||
# 尝试更宽松的匹配
|
||||
topic_pattern = r'"topic":\s*"([^"]+)"[^}]*"contributors":\s*\[([^\]]+)\][^}]*"detail":\s*"([^"]*(?:\\.[^"]*)*)"'
|
||||
matches = re.findall(topic_pattern, result_text, re.DOTALL)
|
||||
|
||||
for match in matches[:max_topics]:
|
||||
topic_name = match[0].strip()
|
||||
contributors_str = match[1].strip()
|
||||
detail = match[2].strip()
|
||||
|
||||
# 清理detail中的转义字符
|
||||
detail = detail.replace('\\"', '"').replace('\\n', ' ').replace('\\t', ' ')
|
||||
|
||||
# 解析参与者列表
|
||||
contributors = []
|
||||
for contrib in re.findall(r'"([^"]+)"', contributors_str):
|
||||
contributors.append(contrib.strip())
|
||||
|
||||
if not contributors:
|
||||
contributors = ["群友"]
|
||||
|
||||
topics.append(SummaryTopic(
|
||||
topic=topic_name,
|
||||
contributors=contributors[:5], # 最多5个参与者
|
||||
detail=detail
|
||||
))
|
||||
|
||||
return topics
|
||||
except Exception as e:
|
||||
logger.error(f"正则表达式提取失败: {e}")
|
||||
return []
|
||||
|
||||
|
||||
async def analyze_user_titles(self, messages: List[Dict], user_analysis: Dict, umo: str = None) -> Tuple[List[UserTitle], TokenUsage]:
|
||||
"""使用LLM分析用户称号"""
|
||||
"""
|
||||
使用LLM分析用户称号
|
||||
保持原有接口,委托给专门的UserTitleAnalyzer处理
|
||||
|
||||
Args:
|
||||
messages: 群聊消息列表
|
||||
user_analysis: 用户分析统计
|
||||
umo: 模型唯一标识符
|
||||
|
||||
Returns:
|
||||
(用户称号列表, Token使用统计)
|
||||
"""
|
||||
try:
|
||||
# 准备用户数据
|
||||
user_summaries = []
|
||||
for user_id, stats in user_analysis.items():
|
||||
if stats["message_count"] < 5: # 过滤活跃度太低的用户
|
||||
continue
|
||||
|
||||
# 分析用户特征
|
||||
night_messages = sum(stats["hours"][h] for h in range(0, 6))
|
||||
day_messages = stats["message_count"] - night_messages
|
||||
avg_chars = stats["char_count"] / stats["message_count"] if stats["message_count"] > 0 else 0
|
||||
|
||||
user_summaries.append({
|
||||
"name": stats["nickname"],
|
||||
"qq": int(user_id),
|
||||
"message_count": stats["message_count"],
|
||||
"avg_chars": round(avg_chars, 1),
|
||||
"emoji_ratio": round(stats["emoji_count"] / stats["message_count"], 2),
|
||||
"night_ratio": round(night_messages / stats["message_count"], 2),
|
||||
"reply_ratio": round(stats["reply_count"] / stats["message_count"], 2)
|
||||
})
|
||||
|
||||
if not user_summaries:
|
||||
return [], TokenUsage()
|
||||
|
||||
# 按消息数量排序,取前N名
|
||||
max_user_titles = self.config_manager.get_max_user_titles()
|
||||
user_summaries.sort(key=lambda x: x["message_count"], reverse=True)
|
||||
user_summaries = user_summaries[:max_user_titles]
|
||||
|
||||
# 构建LLM提示词
|
||||
users_text = "\n".join([
|
||||
f"- {user['name']} (QQ:{user['qq']}): "
|
||||
f"发言{user['message_count']}条, 平均{user['avg_chars']}字, "
|
||||
f"表情比例{user['emoji_ratio']}, 夜间发言比例{user['night_ratio']}, "
|
||||
f"回复比例{user['reply_ratio']}"
|
||||
for user in user_summaries
|
||||
])
|
||||
|
||||
prompt = f"""
|
||||
请为以下群友分配合适的称号和MBTI类型。每个人只能有一个称号,每个称号只能给一个人。
|
||||
|
||||
可选称号:
|
||||
- 龙王: 发言频繁但内容轻松的人
|
||||
- 技术专家: 经常讨论技术话题的人
|
||||
- 夜猫子: 经常在深夜发言的人
|
||||
- 表情包军火库: 经常发表情的人
|
||||
- 沉默终结者: 经常开启话题的人
|
||||
- 评论家: 平均发言长度很长的人
|
||||
- 阳角: 在群里很有影响力的人
|
||||
- 互动达人: 经常回复别人的人
|
||||
- ... (你可以自行进行拓展添加)
|
||||
|
||||
用户数据:
|
||||
{users_text}
|
||||
|
||||
请以JSON格式返回,格式如下:
|
||||
[
|
||||
{{
|
||||
"name": "用户名",
|
||||
"qq": 123456789,
|
||||
"title": "称号",
|
||||
"mbti": "MBTI类型",
|
||||
"reason": "获得此称号的原因"
|
||||
}}
|
||||
]
|
||||
"""
|
||||
|
||||
# 调用LLM
|
||||
response = await self._call_provider_with_retry(None, prompt, max_tokens=1500, temperature=0.5, umo=umo)
|
||||
if response is None:
|
||||
logger.error("用户称号分析调用LLM失败: provider返回None(重试失败)")
|
||||
return [], TokenUsage()
|
||||
|
||||
# 提取token使用统计
|
||||
token_usage = TokenUsage()
|
||||
# 安全地提取 usage,避免 response.raw_completion.usage 为 None 导致的 AttributeError
|
||||
usage = None
|
||||
if getattr(response, 'raw_completion', None) is not None:
|
||||
usage = getattr(response.raw_completion, 'usage', None)
|
||||
if usage:
|
||||
token_usage.prompt_tokens = getattr(usage, 'prompt_tokens', 0) or 0
|
||||
token_usage.completion_tokens = getattr(usage, 'completion_tokens', 0) or 0
|
||||
token_usage.total_tokens = getattr(usage, 'total_tokens', 0) or 0
|
||||
|
||||
# 解析响应
|
||||
if hasattr(response, 'completion_text'):
|
||||
result_text = response.completion_text
|
||||
else:
|
||||
result_text = str(response)
|
||||
|
||||
# debug日志:打印原始响应
|
||||
logger.debug(f"用户称号分析原始响应: {result_text[:500]}...")
|
||||
|
||||
# 尝试解析JSON
|
||||
try:
|
||||
json_match = re.search(r'\[.*\]', result_text, re.DOTALL)
|
||||
if json_match:
|
||||
logger.debug(f"用户称号分析JSON原文: {json_match.group()[:500]}...")
|
||||
titles_data = json.loads(json_match.group())
|
||||
return [UserTitle(**title) for title in titles_data], token_usage
|
||||
except Exception as e:
|
||||
logger.warning(f"用户称号分析JSON解析失败: {e}")
|
||||
logger.warning(f"原始响应: {result_text}")
|
||||
|
||||
return [], token_usage
|
||||
|
||||
logger.info("开始用户称号分析")
|
||||
return await self.user_title_analyzer.analyze_user_titles(messages, user_analysis, umo)
|
||||
except Exception as e:
|
||||
logger.error(f"用户称号分析失败: {e}")
|
||||
return [], TokenUsage()
|
||||
|
||||
|
||||
async def analyze_golden_quotes(self, messages: List[Dict], umo: str = None) -> Tuple[List[GoldenQuote], TokenUsage]:
|
||||
"""使用LLM分析群聊金句"""
|
||||
"""
|
||||
使用LLM分析群聊金句
|
||||
保持原有接口,委托给专门的GoldenQuoteAnalyzer处理
|
||||
|
||||
Args:
|
||||
messages: 群聊消息列表
|
||||
umo: 模型唯一标识符
|
||||
|
||||
Returns:
|
||||
(金句列表, Token使用统计)
|
||||
"""
|
||||
try:
|
||||
# 提取有趣的文本消息
|
||||
interesting_messages = []
|
||||
for msg in messages:
|
||||
sender = msg.get("sender", {})
|
||||
nickname = sender.get("nickname", "") or sender.get("card", "")
|
||||
msg_time = datetime.fromtimestamp(msg.get("time", 0)).strftime("%H:%M")
|
||||
|
||||
for content in msg.get("message", []):
|
||||
if content.get("type") == "text":
|
||||
text = content.get("data", {}).get("text", "").strip()
|
||||
# 过滤长度适中、可能有趣的消息
|
||||
if 5 <= len(text) <= 100 and not text.startswith(("http", "www", "/")):
|
||||
interesting_messages.append({
|
||||
"sender": nickname,
|
||||
"time": msg_time,
|
||||
"content": text
|
||||
})
|
||||
|
||||
if not interesting_messages:
|
||||
return [], TokenUsage()
|
||||
|
||||
# 构建LLM提示词
|
||||
messages_text = "\n".join([
|
||||
f"[{msg['time']}] {msg['sender']}: {msg['content']}"
|
||||
for msg in interesting_messages
|
||||
])
|
||||
|
||||
# 计算金句数量
|
||||
max_golden_quotes = self.config_manager.get_max_golden_quotes()
|
||||
|
||||
prompt = f"""
|
||||
请从以下群聊记录中挑选出{max_golden_quotes}句最具冲击力、最令人惊叹的"金句"。这些金句需满足:
|
||||
- 核心标准:**逆天的神人发言**,即具备颠覆常识的脑洞、逻辑跳脱的表达或强烈反差感的原创内容
|
||||
- 典型特征:包含某些争议话题元素、夸张类比、反常规结论、一本正经的"胡说八道"或突破语境的清奇思路,并且具备一定的冲击力,让人印象深刻。
|
||||
|
||||
对于每个金句,请提供:
|
||||
1. 原文内容(完整保留发言细节)
|
||||
2. 发言人昵称
|
||||
3. 选择理由(具体说明其"逆天"之处,如逻辑颠覆点/脑洞角度/反差感/争议话题元素)
|
||||
|
||||
此外,我将对你进行严格约束:
|
||||
- 优先筛选 **逆天指数最高** 的内容:发情、性压抑话题 > 争议话题 > 元素级 > 颠覆认知级 > 逻辑跳脱级 > 趣味调侃级,剔除单纯玩梗或网络热词堆砌的普通发言
|
||||
- 重点标记包含极端类比、反常识论证或无厘头结论的内容,并且包含一定的争议话题元素。
|
||||
|
||||
群聊记录:
|
||||
{messages_text}
|
||||
|
||||
请以JSON格式返回,格式如下:
|
||||
[
|
||||
{{
|
||||
"content": "金句原文",
|
||||
"sender": "发言人昵称",
|
||||
"reason": "选择这句话的理由(需明确说明逆天特质)"
|
||||
}}
|
||||
]
|
||||
"""
|
||||
|
||||
# 调用LLM
|
||||
response = await self._call_provider_with_retry(None, prompt, max_tokens=1500, temperature=0.7, umo=umo)
|
||||
if response is None:
|
||||
logger.error("金句分析调用LLM失败: provider返回None(重试失败)")
|
||||
return [], TokenUsage()
|
||||
|
||||
# 提取token使用统计
|
||||
token_usage = TokenUsage()
|
||||
# 安全地提取 usage,避免 response.raw_completion.usage 为 None 导致的 AttributeError
|
||||
usage = None
|
||||
if getattr(response, 'raw_completion', None) is not None:
|
||||
usage = getattr(response.raw_completion, 'usage', None)
|
||||
if usage:
|
||||
token_usage.prompt_tokens = getattr(usage, 'prompt_tokens', 0) or 0
|
||||
token_usage.completion_tokens = getattr(usage, 'completion_tokens', 0) or 0
|
||||
token_usage.total_tokens = getattr(usage, 'total_tokens', 0) or 0
|
||||
|
||||
# 解析响应
|
||||
if hasattr(response, 'completion_text'):
|
||||
result_text = response.completion_text
|
||||
else:
|
||||
result_text = str(response)
|
||||
|
||||
# debug日志:打印原始响应
|
||||
logger.debug(f"金句分析原始响应: {result_text[:500]}...")
|
||||
|
||||
# 尝试解析JSON
|
||||
try:
|
||||
json_match = re.search(r'\[.*\]', result_text, re.DOTALL)
|
||||
if json_match:
|
||||
logger.debug(f"金句分析JSON原文: {json_match.group()[:500]}...")
|
||||
quotes_data = json.loads(json_match.group())
|
||||
return [GoldenQuote(**quote) for quote in quotes_data[:max_golden_quotes]], token_usage
|
||||
except Exception as e:
|
||||
logger.warning(f"金句分析JSON解析失败: {e}")
|
||||
logger.warning(f"原始响应: {result_text}")
|
||||
|
||||
return [], token_usage
|
||||
|
||||
logger.info("开始金句分析")
|
||||
return await self.golden_quote_analyzer.analyze_golden_quotes(messages, umo)
|
||||
except Exception as e:
|
||||
logger.error(f"金句分析失败: {e}")
|
||||
return [], TokenUsage()
|
||||
return [], TokenUsage()
|
||||
|
||||
# 向后兼容的方法,保持原有调用方式
|
||||
async def _call_provider_with_retry(self, provider, prompt: str, max_tokens: int,
|
||||
temperature: float, umo: str = None):
|
||||
"""
|
||||
向后兼容的LLM调用方法
|
||||
现在委托给llm_utils模块处理
|
||||
|
||||
Args:
|
||||
provider: LLM服务商实例或None
|
||||
prompt: 输入的提示语
|
||||
max_tokens: 最大生成token数
|
||||
temperature: 采样温度
|
||||
umo: 指定使用的模型唯一标识符
|
||||
|
||||
Returns:
|
||||
LLM生成的结果
|
||||
"""
|
||||
return await call_provider_with_retry(self.context, self.config_manager,
|
||||
prompt, max_tokens, temperature, umo)
|
||||
|
||||
def _fix_json(self, text: str) -> str:
|
||||
"""
|
||||
向后兼容的JSON修复方法
|
||||
现在委托给json_utils模块处理
|
||||
|
||||
Args:
|
||||
text: 需要修复的JSON文本
|
||||
|
||||
Returns:
|
||||
修复后的JSON文本
|
||||
"""
|
||||
return fix_json(text)
|
||||
|
||||
def _extract_topics_with_regex(self, result_text: str, max_topics: int) -> List[SummaryTopic]:
|
||||
"""
|
||||
向后兼容的话题正则提取方法
|
||||
现在委托给json_utils模块处理
|
||||
|
||||
Args:
|
||||
result_text: 需要提取的文本
|
||||
max_topics: 最大话题数量
|
||||
|
||||
Returns:
|
||||
话题对象列表
|
||||
"""
|
||||
|
||||
topics_data = extract_topics_with_regex(result_text, max_topics)
|
||||
return [SummaryTopic(**topic) for topic in topics_data]
|
||||
@@ -0,0 +1,32 @@
|
||||
"""
|
||||
分析工具模块
|
||||
包含JSON处理和LLM API请求处理工具
|
||||
"""
|
||||
|
||||
from .json_utils import (
|
||||
fix_json,
|
||||
parse_json_response,
|
||||
extract_topics_with_regex,
|
||||
extract_user_titles_with_regex,
|
||||
extract_golden_quotes_with_regex
|
||||
)
|
||||
|
||||
from .llm_utils import (
|
||||
call_provider_with_retry,
|
||||
extract_token_usage,
|
||||
extract_response_text
|
||||
)
|
||||
|
||||
__all__ = [
|
||||
# JSON处理工具
|
||||
'fix_json',
|
||||
'parse_json_response',
|
||||
'extract_topics_with_regex',
|
||||
'extract_user_titles_with_regex',
|
||||
'extract_golden_quotes_with_regex',
|
||||
|
||||
# LLM工具
|
||||
'call_provider_with_retry',
|
||||
'extract_token_usage',
|
||||
'extract_response_text'
|
||||
]
|
||||
@@ -0,0 +1,263 @@
|
||||
|
||||
"""
|
||||
JSON处理工具模块
|
||||
提供JSON解析、修复和正则提取功能
|
||||
"""
|
||||
|
||||
import json
|
||||
import re
|
||||
from typing import List, Dict, Tuple, Any, Optional
|
||||
from astrbot.api import logger
|
||||
|
||||
|
||||
def fix_json(text: str) -> str:
|
||||
"""
|
||||
修复JSON格式问题,包括中文符号替换
|
||||
|
||||
Args:
|
||||
text: 需要修复的JSON文本
|
||||
|
||||
Returns:
|
||||
修复后的JSON文本
|
||||
"""
|
||||
try:
|
||||
# 1. 移除markdown代码块标记
|
||||
text = re.sub(r'```json\s*', '', text)
|
||||
text = re.sub(r'```\s*$', '', text)
|
||||
|
||||
# 2. 基础清理
|
||||
text = text.replace('\n', ' ').replace('\r', ' ')
|
||||
text = re.sub(r'\s+', ' ', text)
|
||||
|
||||
# 3. 替换中文符号为英文符号(修复)
|
||||
# 中文引号 -> 英文引号
|
||||
text = text.replace('“', '"').replace('”', '"')
|
||||
text = text.replace('‘', "'").replace('’', "'")
|
||||
# 中文逗号 -> 英文逗号
|
||||
text = text.replace(',', ',')
|
||||
# 中文冒号 -> 英文冒号
|
||||
text = text.replace(':', ':')
|
||||
# 中文括号 -> 英文括号
|
||||
text = text.replace('(', '(').replace(')', ')')
|
||||
text = text.replace('【', '[').replace('】', ']')
|
||||
|
||||
# 4. 处理字符串内容中的特殊字符
|
||||
# 转义字符串内的双引号
|
||||
def escape_quotes_in_strings(match):
|
||||
content = match.group(1)
|
||||
# 转义内部的双引号
|
||||
content = content.replace('"', '\\"')
|
||||
return f'"{content}"'
|
||||
|
||||
# 先处理字段值中的引号
|
||||
text = re.sub(r'"([^"]*(?:"[^"]*)*)"', escape_quotes_in_strings, text)
|
||||
|
||||
# 5. 修复截断的JSON
|
||||
if not text.endswith(']'):
|
||||
last_complete = text.rfind('}')
|
||||
if last_complete > 0:
|
||||
text = text[:last_complete + 1] + ']'
|
||||
|
||||
# 6. 修复常见的JSON格式问题
|
||||
# 1. 修复缺失的逗号
|
||||
text = re.sub(r'}\s*{', '}, {', text)
|
||||
|
||||
# 2. 确保字段名有引号
|
||||
text = re.sub(r'([{,]\s*)([a-zA-Z_][a-zA-Z0-9_]*)\s*:', r'\1"\2":', text)
|
||||
|
||||
# 3. 移除多余的逗号
|
||||
text = re.sub(r',\s*}', '}', text)
|
||||
text = re.sub(r',\s*]', ']', text)
|
||||
|
||||
return text.strip()
|
||||
|
||||
except Exception as e:
|
||||
logger.error(f"JSON修复失败: {e}")
|
||||
return text
|
||||
|
||||
|
||||
def parse_json_response(result_text: str, data_type: str) -> Tuple[bool, Optional[List[Dict]], Optional[str]]:
|
||||
"""
|
||||
统一的JSON解析方法
|
||||
|
||||
Args:
|
||||
result_text: LLM返回的原始文本
|
||||
data_type: 数据类型 ('topics' | 'user_titles' | 'golden_quotes')
|
||||
|
||||
Returns:
|
||||
(成功标志, 解析后的数据列表, 错误消息)
|
||||
"""
|
||||
try:
|
||||
# 1. 提取JSON部分
|
||||
json_match = re.search(r'\[.*?\]', result_text, re.DOTALL)
|
||||
if not json_match:
|
||||
error_msg = f"{data_type}响应中未找到JSON格式"
|
||||
logger.warning(error_msg)
|
||||
return False, None, error_msg
|
||||
|
||||
json_text = json_match.group()
|
||||
logger.debug(f"{data_type}分析JSON原文: {json_text[:500]}...")
|
||||
|
||||
# 2. 修复JSON
|
||||
json_text = fix_json(json_text)
|
||||
logger.debug(f"{data_type}修复后的JSON: {json_text[:300]}...")
|
||||
|
||||
# 3. 解析JSON
|
||||
data = json.loads(json_text)
|
||||
logger.info(f"{data_type}分析成功,解析到 {len(data)} 条数据")
|
||||
return True, data, None
|
||||
|
||||
except json.JSONDecodeError as e:
|
||||
error_msg = f"{data_type}JSON解析失败: {e}"
|
||||
logger.warning(error_msg)
|
||||
logger.debug(f"修复后的JSON: {json_text if 'json_text' in locals() else 'N/A'}")
|
||||
return False, None, error_msg
|
||||
except Exception as e:
|
||||
error_msg = f"{data_type}解析异常: {e}"
|
||||
logger.error(error_msg)
|
||||
return False, None, error_msg
|
||||
|
||||
|
||||
def extract_topics_with_regex(result_text: str, max_topics: int) -> List[Dict]:
|
||||
"""
|
||||
使用正则表达式提取话题信息
|
||||
|
||||
Args:
|
||||
result_text: 需要提取的文本
|
||||
max_topics: 最大话题数量
|
||||
|
||||
Returns:
|
||||
话题数据列表
|
||||
"""
|
||||
try:
|
||||
# 更强的正则表达式提取话题信息,处理转义字符
|
||||
# 匹配每个完整的话题对象
|
||||
topic_pattern = r'\{\s*"topic":\s*"([^"]+)"\s*,\s*"contributors":\s*\[([^\]]+)\]\s*,\s*"detail":\s*"([^"]*(?:\\.[^"]*)*)"\s*\}'
|
||||
matches = re.findall(topic_pattern, result_text, re.DOTALL)
|
||||
|
||||
if not matches:
|
||||
# 尝试更宽松的匹配
|
||||
topic_pattern = r'"topic":\s*"([^"]+)"[^}]*"contributors":\s*\[([^\]]+)\][^}]*"detail":\s*"([^"]*(?:\\.[^"]*)*)"'
|
||||
matches = re.findall(topic_pattern, result_text, re.DOTALL)
|
||||
|
||||
topics = []
|
||||
for match in matches[:max_topics]:
|
||||
topic_name = match[0].strip()
|
||||
contributors_str = match[1].strip()
|
||||
detail = match[2].strip()
|
||||
|
||||
# 清理detail中的转义字符
|
||||
detail = detail.replace('\\"', '"').replace('\\n', ' ').replace('\\t', ' ')
|
||||
|
||||
# 解析参与者列表
|
||||
contributors = []
|
||||
for contrib in re.findall(r'"([^"]+)"', contributors_str):
|
||||
contributors.append(contrib.strip())
|
||||
|
||||
if not contributors:
|
||||
contributors = ["群友"]
|
||||
|
||||
topics.append({
|
||||
"topic": topic_name,
|
||||
"contributors": contributors[:5], # 最多5个参与者
|
||||
"detail": detail
|
||||
})
|
||||
|
||||
return topics
|
||||
|
||||
except Exception as e:
|
||||
logger.error(f"话题正则表达式提取失败: {e}")
|
||||
return []
|
||||
|
||||
|
||||
def extract_user_titles_with_regex(result_text: str, max_count: int) -> List[Dict]:
|
||||
"""
|
||||
使用正则表达式提取用户称号信息
|
||||
|
||||
Args:
|
||||
result_text: 需要提取的文本
|
||||
max_count: 最大提取数量
|
||||
|
||||
Returns:
|
||||
用户称号数据列表
|
||||
"""
|
||||
try:
|
||||
titles = []
|
||||
|
||||
# 正则模式:匹配完整的用户称号对象
|
||||
pattern = r'\{\s*"name":\s*"([^"]+)"\s*,\s*"qq":\s*(\d+)\s*,\s*"title":\s*"([^"]+)"\s*,\s*"mbti":\s*"([^"]+)"\s*,\s*"reason":\s*"([^"]*(?:\\.[^"]*)*)"\s*\}'
|
||||
matches = re.findall(pattern, result_text, re.DOTALL)
|
||||
|
||||
if not matches:
|
||||
# 尝试更宽松的匹配(字段顺序可变)
|
||||
pattern = r'"name":\s*"([^"]+)"[^}]*"qq":\s*(\d+)[^}]*"title":\s*"([^"]+)"[^}]*"mbti":\s*"([^"]+)"[^}]*"reason":\s*"([^"]*(?:\\.[^"]*)*)"'
|
||||
matches = re.findall(pattern, result_text, re.DOTALL)
|
||||
|
||||
for match in matches[:max_count]:
|
||||
name = match[0].strip()
|
||||
qq = int(match[1])
|
||||
title = match[2].strip()
|
||||
mbti = match[3].strip()
|
||||
reason = match[4].strip()
|
||||
|
||||
# 清理转义字符
|
||||
reason = reason.replace('\\"', '"').replace('\\n', ' ').replace('\\t', ' ')
|
||||
|
||||
titles.append({
|
||||
"name": name,
|
||||
"qq": qq,
|
||||
"title": title,
|
||||
"mbti": mbti,
|
||||
"reason": reason
|
||||
})
|
||||
|
||||
return titles
|
||||
|
||||
except Exception as e:
|
||||
logger.error(f"用户称号正则表达式提取失败: {e}")
|
||||
return []
|
||||
|
||||
|
||||
def extract_golden_quotes_with_regex(result_text: str, max_count: int) -> List[Dict]:
|
||||
"""
|
||||
使用正则表达式提取金句信息
|
||||
|
||||
Args:
|
||||
result_text: 需要提取的文本
|
||||
max_count: 最大提取数量
|
||||
|
||||
Returns:
|
||||
金句数据列表
|
||||
"""
|
||||
try:
|
||||
quotes = []
|
||||
|
||||
# 正则模式:匹配完整的金句对象
|
||||
pattern = r'\{\s*"content":\s*"([^"]*(?:\\.[^"]*)*)"\s*,\s*"sender":\s*"([^"]+)"\s*,\s*"reason":\s*"([^"]*(?:\\.[^"]*)*)"\s*\}'
|
||||
matches = re.findall(pattern, result_text, re.DOTALL)
|
||||
|
||||
if not matches:
|
||||
# 尝试更宽松的匹配(字段顺序可变)
|
||||
pattern = r'"content":\s*"([^"]*(?:\\.[^"]*)*)"[^}]*"sender":\s*"([^"]+)"[^}]*"reason":\s*"([^"]*(?:\\.[^"]*)*)"'
|
||||
matches = re.findall(pattern, result_text, re.DOTALL)
|
||||
|
||||
for match in matches[:max_count]:
|
||||
content = match[0].strip()
|
||||
sender = match[1].strip()
|
||||
reason = match[2].strip()
|
||||
|
||||
# 清理转义字符
|
||||
content = content.replace('\\"', '"').replace('\\n', ' ').replace('\\t', ' ')
|
||||
reason = reason.replace('\\"', '"').replace('\\n', ' ').replace('\\t', ' ')
|
||||
|
||||
quotes.append({
|
||||
"content": content,
|
||||
"sender": sender,
|
||||
"reason": reason
|
||||
})
|
||||
|
||||
return quotes
|
||||
|
||||
except Exception as e:
|
||||
logger.error(f"金句正则表达式提取失败: {e}")
|
||||
return []
|
||||
@@ -0,0 +1,172 @@
|
||||
"""
|
||||
LLM API请求处理工具模块
|
||||
提供LLM调用和token统计功能
|
||||
"""
|
||||
|
||||
import asyncio
|
||||
from typing import Optional, Any
|
||||
from astrbot.api import logger
|
||||
import aiohttp
|
||||
|
||||
|
||||
async def call_provider_with_retry(context, config_manager, prompt: str, max_tokens: int,
|
||||
temperature: float, umo: str = None) -> Optional[Any]:
|
||||
"""
|
||||
调用LLM提供者,带超时、重试与退避。支持自定义服务商。
|
||||
|
||||
Args:
|
||||
context: AstrBot上下文对象
|
||||
config_manager: 配置管理器
|
||||
prompt: 输入的提示语
|
||||
max_tokens: 最大生成token数
|
||||
temperature: 采样温度
|
||||
umo: 指定使用的模型唯一标识符
|
||||
|
||||
Returns:
|
||||
LLM生成的结果,失败时返回None
|
||||
"""
|
||||
timeout = config_manager.get_llm_timeout()
|
||||
retries = config_manager.get_llm_retries()
|
||||
backoff = config_manager.get_llm_backoff()
|
||||
|
||||
# 获取自定义服务商参数
|
||||
custom_api_key = config_manager.get_custom_api_key()
|
||||
custom_api_base = config_manager.get_custom_api_base_url()
|
||||
custom_model = config_manager.get_custom_model_name()
|
||||
|
||||
last_exc = None
|
||||
for attempt in range(1, retries + 1):
|
||||
try:
|
||||
if custom_api_key and custom_api_base and custom_model:
|
||||
logger.info(f"使用自定义LLM提供商: {custom_api_base} model={custom_model}")
|
||||
async with aiohttp.ClientSession() as session:
|
||||
headers = {
|
||||
"Authorization": f"Bearer {custom_api_key}",
|
||||
"Content-Type": "application/json"
|
||||
}
|
||||
payload = {
|
||||
"model": custom_model,
|
||||
"messages": [{"role": "user", "content": prompt}],
|
||||
"max_tokens": max_tokens,
|
||||
"temperature": temperature
|
||||
}
|
||||
aio_timeout = aiohttp.ClientTimeout(total=timeout)
|
||||
async with session.post(custom_api_base, json=payload, headers=headers, timeout=aio_timeout) as resp:
|
||||
if resp.status != 200:
|
||||
error_text = await resp.text()
|
||||
logger.error(f"自定义LLM服务商请求失败: HTTP {resp.status}, 内容: {error_text}")
|
||||
try:
|
||||
response_json = await resp.json()
|
||||
except Exception as json_err:
|
||||
error_text = await resp.text()
|
||||
logger.error(f"自定义LLM服务商响应JSON解析失败: {json_err}, 内容: {error_text}")
|
||||
return None
|
||||
# 兼容 OpenAI 格式,安全访问嵌套字段
|
||||
content = None
|
||||
try:
|
||||
choices = response_json.get("choices")
|
||||
if choices and isinstance(choices, list) and len(choices) > 0:
|
||||
message = choices[0].get("message")
|
||||
if message and isinstance(message, dict):
|
||||
content = message.get("content")
|
||||
if content is None:
|
||||
logger.error(f"自定义LLM响应格式异常: {response_json}")
|
||||
return None
|
||||
except Exception as key_err:
|
||||
logger.error(f"自定义LLM响应结构解析失败: {key_err}, 响应内容: {response_json}")
|
||||
return None
|
||||
# 构造一个兼容原有逻辑的对象
|
||||
class CustomResponse:
|
||||
completion_text = content
|
||||
raw_completion = response_json
|
||||
return CustomResponse()
|
||||
else:
|
||||
# 确保使用当前指定的模型
|
||||
provider = context.get_using_provider(umo=umo)
|
||||
provider_id = 'unknown'
|
||||
if provider:
|
||||
try:
|
||||
meta = provider.meta()
|
||||
provider_id = meta.id
|
||||
except Exception as e:
|
||||
logger.debug(f"获取提供商ID失败: {e}")
|
||||
logger.info(f"获取到的 provider ID: {provider_id}")
|
||||
if not provider or provider_id == 'unknown':
|
||||
logger.warning(f"获取的提供商不正确 (Provider ID: {provider_id})")
|
||||
|
||||
logger.info(f"使用LLM provider: {provider}")
|
||||
if not provider:
|
||||
logger.error("provider 为空,无法调用 text_chat,直接返回 None")
|
||||
return None
|
||||
coro = provider.text_chat(prompt=prompt, max_tokens=max_tokens, temperature=temperature)
|
||||
return await asyncio.wait_for(coro, timeout=timeout)
|
||||
except asyncio.TimeoutError as e:
|
||||
last_exc = e
|
||||
logger.warning(f"LLM请求超时: 第{attempt}次, timeout={timeout}s")
|
||||
except Exception as e:
|
||||
last_exc = e
|
||||
logger.warning(f"LLM请求失败: 第{attempt}次, 错误: {last_exc}")
|
||||
# 若非最后一次,等待退避后重试
|
||||
if attempt < retries:
|
||||
await asyncio.sleep(backoff * attempt)
|
||||
|
||||
# 最终仍失败,记录错误并返回 None 由调用方处理降级,避免抛出异常
|
||||
logger.error(f"LLM请求全部重试失败: {last_exc}")
|
||||
return None
|
||||
|
||||
|
||||
def extract_token_usage(response) -> Optional[dict]:
|
||||
"""
|
||||
从LLM响应中提取token使用统计
|
||||
|
||||
Args:
|
||||
response: LLM响应对象
|
||||
|
||||
Returns:
|
||||
Token使用统计字典,包含prompt_tokens, completion_tokens, total_tokens
|
||||
"""
|
||||
try:
|
||||
token_usage = {
|
||||
"prompt_tokens": 0,
|
||||
"completion_tokens": 0,
|
||||
"total_tokens": 0
|
||||
}
|
||||
|
||||
# 安全地提取 usage,避免 response.raw_completion.usage 为 None 导致的 AttributeError
|
||||
usage = None
|
||||
if getattr(response, 'raw_completion', None) is not None:
|
||||
usage = getattr(response.raw_completion, 'usage', None)
|
||||
if usage:
|
||||
token_usage["prompt_tokens"] = getattr(usage, 'prompt_tokens', 0) or 0
|
||||
token_usage["completion_tokens"] = getattr(usage, 'completion_tokens', 0) or 0
|
||||
token_usage["total_tokens"] = getattr(usage, 'total_tokens', 0) or 0
|
||||
|
||||
return token_usage
|
||||
|
||||
except Exception as e:
|
||||
logger.error(f"提取token使用统计失败: {e}")
|
||||
return {
|
||||
"prompt_tokens": 0,
|
||||
"completion_tokens": 0,
|
||||
"total_tokens": 0
|
||||
}
|
||||
|
||||
|
||||
def extract_response_text(response) -> str:
|
||||
"""
|
||||
从LLM响应中提取文本内容
|
||||
|
||||
Args:
|
||||
response: LLM响应对象
|
||||
|
||||
Returns:
|
||||
响应文本内容
|
||||
"""
|
||||
try:
|
||||
if hasattr(response, 'completion_text'):
|
||||
return response.completion_text
|
||||
else:
|
||||
return str(response)
|
||||
except Exception as e:
|
||||
logger.error(f"提取响应文本失败: {e}")
|
||||
return ""
|
||||
Reference in New Issue
Block a user