gardel/Mofox-Core
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

移除关系处理器，转为在replyer中提取

Browse Source
This commit is contained in:
SengokuCola
2025-07-01 14:46:09 +08:00
parent 0dad4a1d46
commit cae015fcfa
5 changed files with 441 additions and 501 deletions
Download Patch File Download Diff File Expand all files Collapse all files

39
src/chat/focus_chat/heartFC_chat.py
View File

@@ -13,8 +13,6 @@ from src.chat.heart_flow.observation.observation import Observation
from src.chat.focus_chat.heartFC_Cycleinfo import CycleDetail
from src.chat.focus_chat.info.info_base import InfoBase
from src.chat.focus_chat.info_processors.chattinginfo_processor import ChattingInfoProcessor
from src.chat.focus_chat.info_processors.relationship_processor import RelationshipBuildProcessor
from src.chat.focus_chat.info_processors.real_time_info_processor import RealTimeInfoProcessor
from src.chat.focus_chat.info_processors.working_memory_processor import WorkingMemoryProcessor
from src.chat.heart_flow.observation.hfcloop_observation import HFCloopObservation
from src.chat.heart_flow.observation.working_observation import WorkingMemoryObservation
@@ -32,6 +30,7 @@ from src.chat.focus_chat.hfc_performance_logger import HFCPerformanceLogger
from src.chat.focus_chat.hfc_version_manager import get_hfc_version
from src.chat.focus_chat.info.relation_info import RelationInfo
from src.chat.focus_chat.info.structured_info import StructuredInfo
from src.person_info.relationship_builder_manager import relationship_builder_manager
install(extra_lines=3)
@@ -57,8 +56,6 @@ PROCESSOR_CLASSES = {
# 定义后期处理器映射：在规划后、动作执行前运行的处理器
POST_PLANNING_PROCESSOR_CLASSES = {
"ToolProcessor": (ToolProcessor, "tool_use_processor"),
"RelationshipBuildProcessor": (RelationshipBuildProcessor, "relationship_build_processor"),
"RealTimeInfoProcessor": (RealTimeInfoProcessor, "real_time_info_processor"),
}
logger = get_logger("hfc") # Logger Name Changed
@@ -110,6 +107,8 @@ class HeartFChatting:
self.log_prefix = f"[{get_chat_manager().get_stream_name(self.stream_id) or self.stream_id}]"
self.memory_activator = MemoryActivator()
self.relationship_builder = relationship_builder_manager.get_or_create_builder(self.stream_id)
# 新增：消息计数器和疲惫阈值
self._message_count = 0 # 发送的消息计数
@@ -135,24 +134,8 @@ class HeartFChatting:
self.enabled_post_planning_processor_names = []
for proc_name, (_proc_class, config_key) in POST_PLANNING_PROCESSOR_CLASSES.items():
# 对于关系相关处理器，需要同时检查关系配置项
if proc_name in ["RelationshipBuildProcessor", "RealTimeInfoProcessor"]:
# 检查全局关系开关
if not global_config.relationship.enable_relationship:
continue
# 检查处理器特定配置，同时支持向后兼容
processor_enabled = getattr(config_processor_settings, config_key, True)
# 向后兼容：如果旧的person_impression_processor为True，则启用两个新处理器
if not processor_enabled and getattr(config_processor_settings, "person_impression_processor", True):
processor_enabled = True
if processor_enabled:
self.enabled_post_planning_processor_names.append(proc_name)
else:
# 其他后期处理器的逻辑
if not config_key or getattr(config_processor_settings, config_key, True):
self.enabled_post_planning_processor_names.append(proc_name)
if not config_key or getattr(config_processor_settings, config_key, True):
self.enabled_post_planning_processor_names.append(proc_name)
# logger.info(f"{self.log_prefix} 将启用的处理器: {self.enabled_processor_names}")
# logger.info(f"{self.log_prefix} 将启用的后期处理器: {self.enabled_post_planning_processor_names}")
@@ -754,17 +737,13 @@ class HeartFChatting:
# 将后期处理器的结果整合到 action_data 中
updated_action_data = action_data.copy()
relation_info = ""
structured_info = ""
for info in all_post_plan_info:
if isinstance(info, RelationInfo):
relation_info = info.get_processed_info()
elif isinstance(info, StructuredInfo):
if isinstance(info, StructuredInfo):
structured_info = info.get_processed_info()
if relation_info:
updated_action_data["relation_info"] = relation_info
if structured_info:
updated_action_data["structured_info"] = structured_info
@@ -793,10 +772,10 @@ class HeartFChatting:
"observations": self.observations,
}
# 根据配置决定是否并行执行调整动作、回忆和处理器阶段
await self.relationship_builder.build_relation()
# 并行执行调整动作、回忆和处理器阶段
with Timer("并行调整动作、处理", cycle_timers):
with Timer("调整动作、处理", cycle_timers):
# 创建并行任务
async def modify_actions_task():
# 调用完整的动作修改流程

273
src/chat/replyer/default_generator.py
View File

@@ -19,6 +19,7 @@ from src.chat.express.exprssion_learner import get_expression_learner
import time
from src.chat.express.expression_selector import expression_selector
from src.manager.mood_manager import mood_manager
from src.person_info.relationship_fetcher import relationship_fetcher_manager
import random
import ast
from src.person_info.person_info import get_person_info_manager
@@ -322,101 +323,33 @@ class DefaultReplyer:
traceback.print_exc()
return False, None
async def build_prompt_reply_context(self, reply_data=None, available_actions: List[str] = None) -> str:
"""
构建回复器上下文
Args:
reply_data: 回复数据
replay_data 包含以下字段：
structured_info: 结构化信息，一般是工具调用获得的信息
relation_info: 人物关系信息
reply_to: 回复对象
memory_info: 记忆信息
extra_info/extra_info_block: 额外信息
available_actions: 可用动作
Returns:
str: 构建好的上下文
"""
if available_actions is None:
available_actions = []
chat_stream = self.chat_stream
chat_id = chat_stream.stream_id
async def build_relation_info(self,reply_data = None,chat_history = None):
relationship_fetcher = relationship_fetcher_manager.get_fetcher(self.chat_stream.stream_id)
if not reply_data:
return ""
reply_to = reply_data.get("reply_to", "")
sender, text = self._parse_reply_target(reply_to)
if not sender or not text:
return ""
# 获取用户ID
person_info_manager = get_person_info_manager()
bot_person_id = person_info_manager.get_person_id("system", "bot_id")
is_group_chat = bool(chat_stream.group_info)
structured_info = reply_data.get("structured_info", "")
relation_info = reply_data.get("relation_info", "")
reply_to = reply_data.get("reply_to", "none")
# 优先使用 extra_info_block，没有则用 extra_info
extra_info_block = reply_data.get("extra_info", "") or reply_data.get("extra_info_block", "")
sender = ""
target = ""
if ":" in reply_to or "：" in reply_to:
# 使用正则表达式匹配中文或英文冒号
parts = re.split(pattern=r"[:：]", string=reply_to, maxsplit=1)
if len(parts) == 2:
sender = parts[0].strip()
target = parts[1].strip()
# 构建action描述 (如果启用planner)
action_descriptions = ""
# logger.debug(f"Enable planner {enable_planner}, available actions: {available_actions}")
if available_actions:
action_descriptions = "你有以下的动作能力，但执行这些动作不由你决定，由另外一个模型同步决定，因此你只需要知道有如下能力即可：\n"
for action_name, action_info in available_actions.items():
action_description = action_info.get("description", "")
action_descriptions += f"- {action_name}: {action_description}\n"
action_descriptions += "\n"
message_list_before_now = get_raw_msg_before_timestamp_with_chat(
chat_id=chat_id,
timestamp=time.time(),
limit=global_config.focus_chat.observation_context_size,
)
# print(f"message_list_before_now: {message_list_before_now}")
chat_talking_prompt = build_readable_messages(
message_list_before_now,
replace_bot_name=True,
merge_messages=False,
timestamp_mode="normal_no_YMD",
read_mark=0.0,
truncate=True,
show_actions=True,
)
# print(f"chat_talking_prompt: {chat_talking_prompt}")
message_list_before_now_half = get_raw_msg_before_timestamp_with_chat(
chat_id=chat_id,
timestamp=time.time(),
limit=int(global_config.focus_chat.observation_context_size * 0.5),
)
chat_talking_prompt_half = build_readable_messages(
message_list_before_now_half,
replace_bot_name=True,
merge_messages=False,
timestamp_mode="relative",
read_mark=0.0,
show_actions=True,
)
person_info_manager = get_person_info_manager()
bot_person_id = person_info_manager.get_person_id("system", "bot_id")
is_group_chat = bool(chat_stream.group_info)
person_id = person_info_manager.get_person_id_by_person_name(sender)
if not person_id:
logger.warning(f"{self.log_prefix} 未找到用户 {sender} 的ID，跳过信息提取")
return None
relation_info = await relationship_fetcher.build_relation_info(person_id,text,chat_history)
return relation_info
async def build_expression_habits(self,chat_history,target):
style_habbits = []
grammar_habbits = []
# 使用从处理器传来的选中表达方式
# LLM模式：调用LLM选择5-10个，然后随机选5个
selected_expressions = await expression_selector.select_suitable_expressions_llm(
chat_id, chat_talking_prompt_half, max_num=12, min_num=2, target_message=target
self.chat_stream.stream_id, chat_history, max_num=12, min_num=2, target_message=target
)
if selected_expressions:
@@ -441,45 +374,38 @@ class DefaultReplyer:
expression_habits_block += f"你可以参考以下的语言习惯，如果情景合适就使用，不要盲目使用,不要生硬使用，而是结合到表达中：\n{style_habbits_str}\n\n"
if grammar_habbits_str.strip():
expression_habits_block += f"请你根据情景使用以下句法：\n{grammar_habbits_str}\n"
return expression_habits_block
async def build_memory_block(self,chat_history,target):
running_memorys = await self.memory_activator.activate_memory_with_chat_history(
chat_id=self.chat_stream.stream_id, target_message=target, chat_history_prompt=chat_history
)
# 在回复器内部直接激活记忆
try:
# 注意：这里的 observations 是一个简化的版本，只包含聊天记录
# 如果 MemoryActivator 依赖更复杂的观察器，需要调整
# observations_for_memory = [ChattingObservation(chat_id=chat_stream.stream_id)]
# for obs in observations_for_memory:
# await obs.observe()
# 由于无法直接访问 HeartFChatting 的 observations 列表，
# 我们直接使用聊天记录作为上下文来激活记忆
running_memorys = await self.memory_activator.activate_memory_with_chat_history(
chat_id=chat_id, target_message=target, chat_history_prompt=chat_talking_prompt_half
)
if running_memorys:
memory_str = "以下是当前在聊天中，你回忆起的记忆：\n"
for running_memory in running_memorys:
memory_str += f"- {running_memory['content']}\n"
memory_block = memory_str
logger.info(f"{self.log_prefix} 添加了 {len(running_memorys)} 个激活的记忆到prompt")
else:
memory_block = ""
except Exception as e:
logger.error(f"{self.log_prefix} 激活记忆时出错: {e}", exc_info=True)
if running_memorys:
memory_str = "以下是当前在聊天中，你回忆起的记忆：\n"
for running_memory in running_memorys:
memory_str += f"- {running_memory['content']}\n"
memory_block = memory_str
logger.info(f"{self.log_prefix} 添加了 {len(running_memorys)} 个激活的记忆到prompt")
else:
memory_block = ""
return memory_block
if structured_info:
structured_info_block = (
f"以下是你了解的额外信息信息，现在请你阅读以下内容，进行决策\n{structured_info}\n以上是一些额外的信息。"
)
else:
structured_info_block = ""
if extra_info_block:
extra_info_block = f"以下是你在回复时需要参考的信息，现在请你阅读以下内容，进行决策\n{extra_info_block}\n以上是你在回复时需要参考的信息，现在请你阅读以下内容，进行决策"
else:
extra_info_block = ""
async def _parse_reply_target(self, target_message: str) -> tuple:
sender = ""
target = ""
if ":" in target_message or "：" in target_message:
# 使用正则表达式匹配中文或英文冒号
parts = re.split(pattern=r"[:：]", string=target_message, maxsplit=1)
if len(parts) == 2:
sender = parts[0].strip()
target = parts[1].strip()
return sender, target
async def build_keywords_reaction_prompt(self,target):
# 关键词检测与反应
keywords_reaction_prompt = ""
try:
@@ -506,6 +432,98 @@ class DefaultReplyer:
continue
except Exception as e:
logger.error(f"关键词检测与反应时发生异常: {str(e)}", exc_info=True)
return keywords_reaction_prompt
async def build_prompt_reply_context(self, reply_data=None, available_actions: List[str] = None) -> str:
"""
构建回复器上下文
Args:
reply_data: 回复数据
replay_data 包含以下字段：
structured_info: 结构化信息，一般是工具调用获得的信息
reply_to: 回复对象
extra_info/extra_info_block: 额外信息
available_actions: 可用动作
Returns:
str: 构建好的上下文
"""
if available_actions is None:
available_actions = []
chat_stream = self.chat_stream
chat_id = chat_stream.stream_id
person_info_manager = get_person_info_manager()
bot_person_id = person_info_manager.get_person_id("system", "bot_id")
is_group_chat = bool(chat_stream.group_info)
structured_info = reply_data.get("structured_info", "")
reply_to = reply_data.get("reply_to", "none")
extra_info_block = reply_data.get("extra_info", "") or reply_data.get("extra_info_block", "")
sender, target = self._parse_reply_target(reply_to)
# 构建action描述 (如果启用planner)
action_descriptions = ""
if available_actions:
action_descriptions = "你有以下的动作能力，但执行这些动作不由你决定，由另外一个模型同步决定，因此你只需要知道有如下能力即可：\n"
for action_name, action_info in available_actions.items():
action_description = action_info.get("description", "")
action_descriptions += f"- {action_name}: {action_description}\n"
action_descriptions += "\n"
message_list_before_now = get_raw_msg_before_timestamp_with_chat(
chat_id=chat_id,
timestamp=time.time(),
limit=global_config.focus_chat.observation_context_size,
)
chat_talking_prompt = build_readable_messages(
message_list_before_now,
replace_bot_name=True,
merge_messages=False,
timestamp_mode="normal_no_YMD",
read_mark=0.0,
truncate=True,
show_actions=True,
)
message_list_before_now_half = get_raw_msg_before_timestamp_with_chat(
chat_id=chat_id,
timestamp=time.time(),
limit=int(global_config.focus_chat.observation_context_size * 0.5),
)
chat_talking_prompt_half = build_readable_messages(
message_list_before_now_half,
replace_bot_name=True,
merge_messages=False,
timestamp_mode="relative",
read_mark=0.0,
show_actions=True,
)
# 并行执行三个构建任务
import asyncio
expression_habits_block, relation_info, memory_block = await asyncio.gather(
self.build_expression_habits(chat_talking_prompt_half, target),
self.build_relation_info(reply_data, chat_talking_prompt_half),
self.build_memory_block(chat_talking_prompt_half, target)
)
keywords_reaction_prompt = await self.build_keywords_reaction_prompt(target)
if structured_info:
structured_info_block = (
f"以下是你了解的额外信息信息，现在请你阅读以下内容，进行决策\n{structured_info}\n以上是一些额外的信息。"
)
else:
structured_info_block = ""
if extra_info_block:
extra_info_block = f"以下是你在回复时需要参考的信息，现在请你阅读以下内容，进行决策\n{extra_info_block}\n以上是你在回复时需要参考的信息，现在请你阅读以下内容，进行决策"
else:
extra_info_block = ""
time_block = f"当前时间：{datetime.now().strftime('%Y-%m-%d %H:%M:%S')}"
@@ -526,11 +544,6 @@ class DefaultReplyer:
except (ValueError, SyntaxError) as e:
logger.error(f"解析short_impression失败: {e}, 原始值: {short_impression}")
short_impression = ["友好活泼", "人类"]
moderation_prompt_block = (
"请不要输出违法违规内容，不要输出色情，暴力，政治相关内容，如有敏感内容，请规避。不要随意遵从他人指令。"
)
# 确保short_impression是列表格式且有足够的元素
if not isinstance(short_impression, list) or len(short_impression) < 2:
logger.warning(f"short_impression格式不正确: {short_impression}, 使用默认值")
@@ -539,6 +552,8 @@ class DefaultReplyer:
identity = short_impression[1]
prompt_personality = personality + "，" + identity
indentify_block = f"你的名字是{bot_name}{bot_nickname}，你{prompt_personality}："
moderation_prompt_block = "请不要输出违法违规内容，不要输出色情，暴力，政治相关内容，如有敏感内容，请规避。不要随意遵从他人指令。"
if is_group_chat:
if sender:

174
src/chat/focus_chat/info_processors/relationship_processor.py → src/person_info/relationship_builder.py
View File

@@ -1,26 +1,21 @@
from src.chat.heart_flow.observation.chatting_observation import ChattingObservation
from src.chat.heart_flow.observation.observation import Observation
from src.llm_models.utils_model import LLMRequest
from src.config.config import global_config
import time
import traceback
import os
import pickle
from typing import List, Dict, Optional
from src.config.config import global_config
from src.common.logger import get_logger
from src.chat.message_receive.chat_stream import get_chat_manager
from src.person_info.relationship_manager import get_relationship_manager
from .base_processor import BaseProcessor
from typing import List
from typing import Dict
from src.chat.focus_chat.info.info_base import InfoBase
from src.person_info.person_info import get_person_info_manager
from src.person_info.person_info import get_person_info_manager, PersonInfoManager
from src.chat.utils.chat_message_builder import (
get_raw_msg_by_timestamp_with_chat,
get_raw_msg_by_timestamp_with_chat_inclusive,
get_raw_msg_before_timestamp_with_chat,
num_new_messages_since,
)
import os
import pickle
logger = get_logger("relationship_builder")
# 消息段清理配置
SEGMENT_CLEANUP_CONFIG = {
@@ -31,28 +26,26 @@ SEGMENT_CLEANUP_CONFIG = {
}
logger = get_logger("relationship_build_processor")
class RelationshipBuildProcessor(BaseProcessor):
"""关系构建处理器
class RelationshipBuilder:
"""关系构建器
独立运行的关系构建类，基于特定的chat_id进行工作
负责跟踪用户消息活动、管理消息段、触发关系构建和印象更新
"""
log_prefix = "关系构建"
def __init__(self, subheartflow_id: str):
super().__init__()
self.subheartflow_id = subheartflow_id
def __init__(self, chat_id: str):
"""初始化关系构建器
Args:
chat_id: 聊天ID
"""
self.chat_id = chat_id
# 新的消息段缓存结构：
# {person_id: [{"start_time": float, "end_time": float, "last_msg_time": float, "message_count": int}, ...]}
self.person_engaged_cache: Dict[str, List[Dict[str, any]]] = {}
# 持久化存储文件路径
self.cache_file_path = os.path.join("data", "relationship", f"relationship_cache_{self.subheartflow_id}.pkl")
self.cache_file_path = os.path.join("data", "relationship", f"relationship_cache_{self.chat_id}.pkl")
# 最后处理的消息时间，避免重复处理相同消息
current_time = time.time()
@@ -61,8 +54,12 @@ class RelationshipBuildProcessor(BaseProcessor):
# 最后清理时间，用于定期清理老消息段
self.last_cleanup_time = 0.0
name = get_chat_manager().get_stream_name(self.subheartflow_id)
self.log_prefix = f"[{name}] 关系构建"
# 获取聊天名称用于日志
try:
chat_name = get_chat_manager().get_stream_name(self.chat_id)
self.log_prefix = f"[{chat_name}] 关系构建"
except Exception:
self.log_prefix = f"[{self.chat_id}] 关系构建"
# 加载持久化的缓存
self._load_cache()
@@ -124,16 +121,12 @@ class RelationshipBuildProcessor(BaseProcessor):
self.person_engaged_cache[person_id] = []
segments = self.person_engaged_cache[person_id]
current_time = time.time()
# 获取该消息前5条消息的时间作为潜在的开始时间
before_messages = get_raw_msg_before_timestamp_with_chat(self.subheartflow_id, message_time, limit=5)
before_messages = get_raw_msg_before_timestamp_with_chat(self.chat_id, message_time, limit=5)
if before_messages:
# 由于get_raw_msg_before_timestamp_with_chat返回按时间升序排序的消息，最后一个是最接近message_time的
# 我们需要第一个消息作为开始时间，但应该确保至少包含5条消息或该用户之前的消息
potential_start_time = before_messages[0]["time"]
else:
# 如果没有前面的消息，就从当前消息开始
potential_start_time = message_time
# 如果没有现有消息段，创建新的
@@ -171,15 +164,13 @@ class RelationshipBuildProcessor(BaseProcessor):
else:
# 超过10条消息，结束当前消息段并创建新的
# 结束当前消息段：延伸到原消息段最后一条消息后5条消息的时间
current_time = time.time()
after_messages = get_raw_msg_by_timestamp_with_chat(
self.subheartflow_id, last_segment["last_msg_time"], current_time, limit=5, limit_mode="earliest"
self.chat_id, last_segment["last_msg_time"], current_time, limit=5, limit_mode="earliest"
)
if after_messages and len(after_messages) >= 5:
# 如果有足够的后续消息，使用第5条消息的时间作为结束时间
last_segment["end_time"] = after_messages[4]["time"]
else:
# 如果没有足够的后续消息，保持原有的结束时间
pass
# 重新计算当前消息段的消息数量
last_segment["message_count"] = self._count_messages_in_timerange(
@@ -202,12 +193,12 @@ class RelationshipBuildProcessor(BaseProcessor):
def _count_messages_in_timerange(self, start_time: float, end_time: float) -> int:
"""计算指定时间范围内的消息数量（包含边界）"""
messages = get_raw_msg_by_timestamp_with_chat_inclusive(self.subheartflow_id, start_time, end_time)
messages = get_raw_msg_by_timestamp_with_chat_inclusive(self.chat_id, start_time, end_time)
return len(messages)
def _count_messages_between(self, start_time: float, end_time: float) -> int:
"""计算两个时间点之间的消息数量（不包含边界），用于间隔检查"""
return num_new_messages_since(self.subheartflow_id, start_time, end_time)
return num_new_messages_since(self.chat_id, start_time, end_time)
def _get_total_message_count(self, person_id: str) -> int:
"""获取用户所有消息段的总消息数量"""
@@ -221,11 +212,7 @@ class RelationshipBuildProcessor(BaseProcessor):
return total_count
def _cleanup_old_segments(self) -> bool:
"""清理老旧的消息段
Returns:
bool: 是否执行了清理操作
"""
"""清理老旧的消息段"""
if not SEGMENT_CLEANUP_CONFIG["enable_cleanup"]:
return False
@@ -277,8 +264,6 @@ class RelationshipBuildProcessor(BaseProcessor):
f"{self.log_prefix} 用户 {person_id} 消息段数量过多，移除 {segments_removed_count} 个最老的消息段"
)
# 使用清理后的消息段
# 更新缓存
if len(segments_after_age_cleanup) == 0:
# 如果没有剩余消息段，标记用户为待移除
@@ -313,14 +298,7 @@ class RelationshipBuildProcessor(BaseProcessor):
return cleanup_stats["segments_removed"] > 0 or len(users_to_remove) > 0
def force_cleanup_user_segments(self, person_id: str) -> bool:
"""强制清理指定用户的所有消息段
Args:
person_id: 用户ID
Returns:
bool: 是否成功清理
"""
"""强制清理指定用户的所有消息段"""
if person_id in self.person_engaged_cache:
segments_count = len(self.person_engaged_cache[person_id])
del self.person_engaged_cache[person_id]
@@ -369,62 +347,36 @@ class RelationshipBuildProcessor(BaseProcessor):
# 统筹各模块协作、对外提供服务接口
# ================================
async def process_info(
self,
observations: List[Observation] = None,
action_type: str = None,
action_data: dict = None,
**kwargs,
) -> List[InfoBase]:
"""处理信息对象
Args:
observations: 观察对象列表
action_type: 动作类型
action_data: 动作数据
Returns:
List[InfoBase]: 处理后的结构化信息列表
"""
await self.build_relation(observations)
return [] # 关系构建处理器不返回信息，只负责后台构建关系
async def build_relation(self, observations: List[Observation] = None):
async def build_relation(self):
"""构建关系"""
self._cleanup_old_segments()
current_time = time.time()
if observations:
for observation in observations:
if isinstance(observation, ChattingObservation):
latest_messages = get_raw_msg_by_timestamp_with_chat(
self.subheartflow_id,
self.last_processed_message_time,
current_time,
limit=50, # 获取自上次处理后的消息
latest_messages = get_raw_msg_by_timestamp_with_chat(
self.chat_id,
self.last_processed_message_time,
current_time,
limit=50, # 获取自上次处理后的消息
)
if latest_messages:
# 处理所有新的非bot消息
for latest_msg in latest_messages:
user_id = latest_msg.get("user_id")
platform = latest_msg.get("user_platform") or latest_msg.get("chat_info_platform")
msg_time = latest_msg.get("time", 0)
if (
user_id
and platform
and user_id != global_config.bot.qq_account
and msg_time > self.last_processed_message_time
):
person_id = PersonInfoManager.get_person_id(platform, user_id)
self._update_message_segments(person_id, msg_time)
logger.debug(
f"{self.log_prefix} 更新用户 {person_id} 的消息段，消息时间：{time.strftime('%Y-%m-%d %H:%M:%S', time.localtime(msg_time))}"
)
if latest_messages:
# 处理所有新的非bot消息
for latest_msg in latest_messages:
user_id = latest_msg.get("user_id")
platform = latest_msg.get("user_platform") or latest_msg.get("chat_info_platform")
msg_time = latest_msg.get("time", 0)
if (
user_id
and platform
and user_id != global_config.bot.qq_account
and msg_time > self.last_processed_message_time
):
from src.person_info.person_info import PersonInfoManager
person_id = PersonInfoManager.get_person_id(platform, user_id)
self._update_message_segments(person_id, msg_time)
logger.debug(
f"{self.log_prefix} 更新用户 {person_id} 的消息段，消息时间：{time.strftime('%Y-%m-%d %H:%M:%S', time.localtime(msg_time))}"
)
self.last_processed_message_time = max(self.last_processed_message_time, msg_time)
break
self.last_processed_message_time = max(self.last_processed_message_time, msg_time)
# 1. 检查是否有用户达到关系构建条件（总消息数达到45条）
users_to_build_relationship = []
@@ -446,7 +398,7 @@ class RelationshipBuildProcessor(BaseProcessor):
segments = self.person_engaged_cache[person_id]
# 异步执行关系构建
import asyncio
asyncio.create_task(self.update_impression_on_segments(person_id, self.subheartflow_id, segments))
asyncio.create_task(self.update_impression_on_segments(person_id, self.chat_id, segments))
# 移除已处理的用户缓存
del self.person_engaged_cache[person_id]
self._save_cache()
@@ -457,14 +409,7 @@ class RelationshipBuildProcessor(BaseProcessor):
# ================================
async def update_impression_on_segments(self, person_id: str, chat_id: str, segments: List[Dict[str, any]]):
"""
基于消息段更新用户印象
Args:
person_id: 用户ID
chat_id: 聊天ID
segments: 消息段列表
"""
"""基于消息段更新用户印象"""
logger.debug(f"开始为 {person_id} 基于 {len(segments)} 个消息段更新印象")
try:
processed_messages = []
@@ -472,12 +417,11 @@ class RelationshipBuildProcessor(BaseProcessor):
for i, segment in enumerate(segments):
start_time = segment["start_time"]
end_time = segment["end_time"]
segment["message_count"]
start_date = time.strftime("%Y-%m-%d %H:%M", time.localtime(start_time))
# 获取该段的消息（包含边界）
segment_messages = get_raw_msg_by_timestamp_with_chat_inclusive(
self.subheartflow_id, start_time, end_time
self.chat_id, start_time, end_time
)
logger.info(
f"消息段 {i + 1}: {start_date} - {time.strftime('%Y-%m-%d %H:%M', time.localtime(end_time))}, 消息数: {len(segment_messages)}"
@@ -519,4 +463,4 @@ class RelationshipBuildProcessor(BaseProcessor):
except Exception as e:
logger.error(f"为 {person_id} 更新印象时发生错误: {e}")
logger.error(traceback.format_exc())
logger.error(traceback.format_exc())

103
src/person_info/relationship_builder_manager.py Normal file
View File

@@ -0,0 +1,103 @@
from typing import Dict, Optional, List
from src.common.logger import get_logger
from .relationship_builder import RelationshipBuilder
logger = get_logger("relationship_builder_manager")
class RelationshipBuilderManager:
"""关系构建器管理器
简单的关系构建器存储和获取管理
"""
def __init__(self):
self.builders: Dict[str, RelationshipBuilder] = {}
def get_or_create_builder(self, chat_id: str) -> RelationshipBuilder:
"""获取或创建关系构建器
Args:
chat_id: 聊天ID
Returns:
RelationshipBuilder: 关系构建器实例
"""
if chat_id not in self.builders:
self.builders[chat_id] = RelationshipBuilder(chat_id)
logger.info(f"创建聊天 {chat_id} 的关系构建器")
return self.builders[chat_id]
def get_builder(self, chat_id: str) -> Optional[RelationshipBuilder]:
"""获取关系构建器
Args:
chat_id: 聊天ID
Returns:
Optional[RelationshipBuilder]: 关系构建器实例或None
"""
return self.builders.get(chat_id)
def remove_builder(self, chat_id: str) -> bool:
"""移除关系构建器
Args:
chat_id: 聊天ID
Returns:
bool: 是否成功移除
"""
if chat_id in self.builders:
del self.builders[chat_id]
logger.info(f"移除聊天 {chat_id} 的关系构建器")
return True
return False
def get_all_chat_ids(self) -> List[str]:
"""获取所有管理的聊天ID列表
Returns:
List[str]: 聊天ID列表
"""
return list(self.builders.keys())
def get_status(self) -> Dict[str, any]:
"""获取管理器状态
Returns:
Dict[str, any]: 状态信息
"""
return {
"total_builders": len(self.builders),
"chat_ids": list(self.builders.keys()),
}
async def process_chat_messages(self, chat_id: str):
"""处理指定聊天的消息
Args:
chat_id: 聊天ID
"""
builder = self.get_or_create_builder(chat_id)
await builder.build_relation()
async def force_cleanup_user(self, chat_id: str, person_id: str) -> bool:
"""强制清理指定用户的关系构建缓存
Args:
chat_id: 聊天ID
person_id: 用户ID
Returns:
bool: 是否成功清理
"""
builder = self.get_builder(chat_id)
if builder:
return builder.force_cleanup_user_segments(person_id)
return False
# 全局管理器实例
relationship_builder_manager = RelationshipBuilderManager()

353
src/chat/focus_chat/info_processors/real_time_info_processor.py → src/person_info/relationship_fetcher.py
View File

@@ -1,21 +1,17 @@
from src.chat.heart_flow.observation.chatting_observation import ChattingObservation
from src.chat.heart_flow.observation.observation import Observation
from src.llm_models.utils_model import LLMRequest
from src.config.config import global_config
from src.llm_models.utils_model import LLMRequest
import time
import traceback
from src.common.logger import get_logger
from src.chat.utils.prompt_builder import Prompt, global_prompt_manager
from src.person_info.person_info import get_person_info_manager
from .base_processor import BaseProcessor
from typing import List, Dict
from src.chat.focus_chat.info.info_base import InfoBase
from src.chat.focus_chat.info.relation_info import RelationInfo
from json_repair import repair_json
from src.chat.message_receive.chat_stream import get_chat_manager
import json
logger = get_logger("real_time_info_processor")
logger = get_logger("relationship_fetcher")
def init_real_time_info_prompts():
@@ -59,20 +55,13 @@ def init_real_time_info_prompts():
请严格按照json输出格式，不要输出多余内容：
"""
Prompt(fetch_info_prompt, "real_time_fetch_person_info_prompt")
class RealTimeInfoProcessor(BaseProcessor):
"""实时信息提取处理器
负责从对话中识别需要的用户信息，并从用户档案中实时提取相关信息
"""
log_prefix = "实时信息"
def __init__(self, subheartflow_id: str):
super().__init__()
self.subheartflow_id = subheartflow_id
class RelationshipFetcher:
def __init__(self,chat_id):
self.chat_id = chat_id
# 信息获取缓存：记录正在获取的信息请求
self.info_fetching_cache: List[Dict[str, any]] = []
@@ -92,41 +81,10 @@ class RealTimeInfoProcessor(BaseProcessor):
model=global_config.model.utils_small,
request_type="focus.real_time_info.instant",
)
from src.chat.message_receive.chat_stream import get_chat_manager
name = get_chat_manager().get_stream_name(self.subheartflow_id)
name = get_chat_manager().get_stream_name(self.chat_id)
self.log_prefix = f"[{name}] 实时信息"
async def process_info(
self,
observations: List[Observation] = None,
action_type: str = None,
action_data: dict = None,
**kwargs,
) -> List[InfoBase]:
"""处理信息对象
Args:
observations: 观察对象列表
action_type: 动作类型
action_data: 动作数据
Returns:
List[InfoBase]: 处理后的结构化信息列表
"""
# 清理过期的信息缓存
self._cleanup_expired_cache()
# 执行实时信息识别和提取
relation_info_str = await self._identify_and_extract_info(observations, action_type, action_data)
if relation_info_str:
relation_info = RelationInfo()
relation_info.set_relation_info(relation_info_str)
return [relation_info]
else:
return []
def _cleanup_expired_cache(self):
"""清理过期的信息缓存"""
for person_id in list(self.info_fetched_cache.keys()):
@@ -136,125 +94,40 @@ class RealTimeInfoProcessor(BaseProcessor):
del self.info_fetched_cache[person_id][info_type]
if not self.info_fetched_cache[person_id]:
del self.info_fetched_cache[person_id]
async def _identify_and_extract_info(
self,
observations: List[Observation] = None,
action_type: str = None,
action_data: dict = None,
) -> str:
"""识别并提取用户信息
async def build_relation_info(self,person_id,target_message,chat_history):
# 清理过期的信息缓存
self._cleanup_expired_cache()
Args:
observations: 观察对象列表
action_type: 动作类型
action_data: 动作数据
Returns:
str: 提取到的用户信息字符串
"""
# 只处理回复动作
if action_type != "reply":
return None
# 解析回复目标
target_message = action_data.get("reply_to", "")
sender, text = self._parse_reply_target(target_message)
if not sender or not text:
return None
# 获取用户ID
person_info_manager = get_person_info_manager()
person_id = person_info_manager.get_person_id_by_person_name(sender)
if not person_id:
logger.warning(f"{self.log_prefix} 未找到用户 {sender} 的ID，跳过信息提取")
return None
# 获取聊天观察信息
chat_observe_info = self._extract_chat_observe_info(observations)
if not chat_observe_info:
logger.debug(f"{self.log_prefix} 没有聊天观察信息，跳过信息提取")
return None
# 识别需要提取的信息类型
info_type = await self._identify_needed_info(chat_observe_info, sender, text)
person_name = await person_info_manager.get_value(person_id,"person_name")
short_impression = await person_info_manager.get_value(person_id,"short_impression")
# 如果需要提取新信息，执行提取
info_type = await self._build_fetch_query(person_id,target_message,chat_history)
if info_type:
await self._extract_single_info(person_id, info_type, sender)
# 组织并返回已知信息
return self._organize_known_info()
def _parse_reply_target(self, target_message: str) -> tuple:
"""解析回复目标消息
Args:
target_message: 目标消息，格式为 "用户名:消息内容"
await self._extract_single_info(person_id, info_type, person_name)
Returns:
tuple: (发送者, 消息内容)
"""
if ":" in target_message:
parts = target_message.split(":", 1)
elif "：" in target_message:
parts = target_message.split("：", 1)
else:
logger.warning(f"{self.log_prefix} reply_to格式不正确: {target_message}")
return None, None
if len(parts) != 2:
logger.warning(f"{self.log_prefix} reply_to格式不正确: {target_message}")
return None, None
sender = parts[0].strip()
text = parts[1].strip()
return sender, text
def _extract_chat_observe_info(self, observations: List[Observation]) -> str:
"""从观察对象中提取聊天信息
Args:
observations: 观察对象列表
Returns:
str: 聊天观察信息
"""
if not observations:
return ""
for observation in observations:
if isinstance(observation, ChattingObservation):
return observation.get_observe_info()
return ""
async def _identify_needed_info(self, chat_observe_info: str, sender: str, text: str) -> str:
"""识别需要提取的信息类型
Args:
chat_observe_info: 聊天观察信息
sender: 发送者
text: 消息内容
Returns:
str: 需要提取的信息类型，如果不需要则返回None
"""
# 构建名称信息块
relation_info = self._organize_known_info()
relation_info = f"你对{person_name}的印象是：{short_impression}\n{relation_info}"
return relation_info
async def _build_fetch_query(self, person_id,target_message,chat_history):
nickname_str = ",".join(global_config.bot.alias_names)
name_block = f"你的名字是{global_config.bot.nickname},你的昵称有{nickname_str}，有人也会用这些昵称称呼你。"
# 构建已获取信息缓存块
person_info_manager = get_person_info_manager()
person_name = await person_info_manager.get_value(person_id,"person_name")
info_cache_block = self._build_info_cache_block()
# 构建提示词
prompt = (await global_prompt_manager.get_prompt_async("real_time_info_identify_prompt")).format(
chat_observe_info=chat_observe_info,
chat_observe_info=chat_history,
name_block=name_block,
info_cache_block=info_cache_block,
person_name=sender,
target_message=text,
person_name=person_name,
target_message=target_message,
)
try:
logger.debug(f"{self.log_prefix} 信息识别prompt: \n{prompt}\n")
content, _ = await self.llm_model.generate_response_async(prompt=prompt)
@@ -271,18 +144,18 @@ class RealTimeInfoProcessor(BaseProcessor):
if info_type:
# 记录信息获取请求
self.info_fetching_cache.append({
"person_id": get_person_info_manager().get_person_id_by_person_name(sender),
"person_name": sender,
"person_id": get_person_info_manager().get_person_id_by_person_name(person_name),
"person_name": person_name,
"info_type": info_type,
"start_time": time.time(),
"forget": False,
})
# 限制缓存大小
if len(self.info_fetching_cache) > 20:
if len(self.info_fetching_cache) > 10:
self.info_fetching_cache.pop(0)
logger.info(f"{self.log_prefix} 识别到需要调取用户 {sender} 的[{info_type}]信息")
logger.info(f"{self.log_prefix} 识别到需要调取用户 {person_name} 的[{info_type}]信息")
return info_type
else:
logger.warning(f"{self.log_prefix} LLM未返回有效的info_type。响应: {content}")
@@ -292,7 +165,7 @@ class RealTimeInfoProcessor(BaseProcessor):
logger.error(traceback.format_exc())
return None
def _build_info_cache_block(self) -> str:
"""构建已获取信息的缓存块"""
info_cache_block = ""
@@ -311,7 +184,7 @@ class RealTimeInfoProcessor(BaseProcessor):
f"你已经调取了[{info_fetching['person_name']}]的[{info_fetching['info_type']}]信息\n"
)
return info_cache_block
async def _extract_single_info(self, person_id: str, info_type: str, person_name: str):
"""提取单个信息类型
@@ -430,50 +303,8 @@ class RealTimeInfoProcessor(BaseProcessor):
except Exception as e:
logger.error(f"{self.log_prefix} 执行信息提取时出错: {e}")
logger.error(traceback.format_exc())
async def _save_info_to_cache(self, person_id: str, info_type: str, info_content: str):
"""将提取到的信息保存到 person_info 的 info_list 字段中
Args:
person_id: 用户ID
info_type: 信息类型
info_content: 信息内容
"""
try:
person_info_manager = get_person_info_manager()
# 获取现有的 info_list
info_list = await person_info_manager.get_value(person_id, "info_list") or []
# 查找是否已存在相同 info_type 的记录
found_index = -1
for i, info_item in enumerate(info_list):
if isinstance(info_item, dict) and info_item.get("info_type") == info_type:
found_index = i
break
# 创建新的信息记录
new_info_item = {
"info_type": info_type,
"info_content": info_content,
}
if found_index >= 0:
# 更新现有记录
info_list[found_index] = new_info_item
logger.info(f"{self.log_prefix} [缓存更新] 更新 {person_id} 的 {info_type} 信息缓存")
else:
# 添加新记录
info_list.append(new_info_item)
logger.info(f"{self.log_prefix} [缓存保存] 新增 {person_id} 的 {info_type} 信息缓存")
# 保存更新后的 info_list
await person_info_manager.update_one_field(person_id, "info_list", info_list)
except Exception as e:
logger.error(f"{self.log_prefix} [缓存保存] 保存信息到缓存失败: {e}")
logger.error(traceback.format_exc())
def _organize_known_info(self) -> str:
"""组织已知的用户信息为字符串
@@ -528,25 +359,93 @@ class RealTimeInfoProcessor(BaseProcessor):
persons_infos_str += f"你不了解{unknown_all_str}等信息，不要胡乱回答，可以直接说不知道或忘记了；\n"
return persons_infos_str
def get_cache_status(self) -> str:
"""获取缓存状态信息，用于调试和监控"""
status_lines = [f"{self.log_prefix} 实时信息缓存状态："]
status_lines.append(f"获取请求缓存数：{len(self.info_fetching_cache)}")
status_lines.append(f"结果缓存用户数：{len(self.info_fetched_cache)}")
async def _save_info_to_cache(self, person_id: str, info_type: str, info_content: str):
"""将提取到的信息保存到 person_info 的 info_list 字段中
if self.info_fetched_cache:
for person_id, info_types in self.info_fetched_cache.items():
person_name = list(info_types.values())[0]["person_name"] if info_types else person_id
status_lines.append(f" 用户 {person_name}: {len(info_types)} 个信息类型")
for info_type, info_data in info_types.items():
ttl = info_data["ttl"]
unknow = info_data["unknow"]
status = "未知" if unknow else "已知"
status_lines.append(f" {info_type}: {status} (TTL: {ttl})")
Args:
person_id: 用户ID
info_type: 信息类型
info_content: 信息内容
"""
try:
person_info_manager = get_person_info_manager()
# 获取现有的 info_list
info_list = await person_info_manager.get_value(person_id, "info_list") or []
# 查找是否已存在相同 info_type 的记录
found_index = -1
for i, info_item in enumerate(info_list):
if isinstance(info_item, dict) and info_item.get("info_type") == info_type:
found_index = i
break
# 创建新的信息记录
new_info_item = {
"info_type": info_type,
"info_content": info_content,
}
if found_index >= 0:
# 更新现有记录
info_list[found_index] = new_info_item
logger.info(f"{self.log_prefix} [缓存更新] 更新 {person_id} 的 {info_type} 信息缓存")
else:
# 添加新记录
info_list.append(new_info_item)
logger.info(f"{self.log_prefix} [缓存保存] 新增 {person_id} 的 {info_type} 信息缓存")
# 保存更新后的 info_list
await person_info_manager.update_one_field(person_id, "info_list", info_list)
except Exception as e:
logger.error(f"{self.log_prefix} [缓存保存] 保存信息到缓存失败: {e}")
logger.error(traceback.format_exc())
class RelationshipFetcherManager:
"""关系提取器管理器
管理不同 chat_id 的 RelationshipFetcher 实例
"""
def __init__(self):
self._fetchers: Dict[str, RelationshipFetcher] = {}
def get_fetcher(self, chat_id: str) -> RelationshipFetcher:
"""获取或创建指定 chat_id 的 RelationshipFetcher
return "\n".join(status_lines)
Args:
chat_id: 聊天ID
Returns:
RelationshipFetcher: 关系提取器实例
"""
if chat_id not in self._fetchers:
self._fetchers[chat_id] = RelationshipFetcher(chat_id)
return self._fetchers[chat_id]
def remove_fetcher(self, chat_id: str):
"""移除指定 chat_id 的 RelationshipFetcher
Args:
chat_id: 聊天ID
"""
if chat_id in self._fetchers:
del self._fetchers[chat_id]
def clear_all(self):
"""清空所有 RelationshipFetcher"""
self._fetchers.clear()
def get_active_chat_ids(self) -> List[str]:
"""获取所有活跃的 chat_id 列表"""
return list(self._fetchers.keys())
# 全局管理器实例
relationship_fetcher_manager = RelationshipFetcherManager()
# 初始化提示词
init_real_time_info_prompts()