feat: 重构成员画像为日周月分层沉淀链路并增强后台摘要能力
本次提交围绕成员画像插件进行了较大升级，核心目标是把原来偏单次、偏近期的成员交互摘要，升级为可随时间沉淀的分层画像能力。 主要功能变更如下： 1. 新增成员分层摘要存储表 t_member_digest，并提供对应的数据库操作层，支持按成员、按群、按摘要类型（daily/weekly/monthly）持久化周期性摘要结果。 2. 在 member_context 插件内新增 MemberDigestService，把画像生成拆分为日摘要、周摘要、月摘要三级处理流程，再由最终画像服务消费这些分层摘要，减少直接反复处理大量原始消息带来的成本和失真。 3. 新增提示词构建模块，分别为日级观察、周级归纳、月级归纳以及最终画像整理提供独立提示词，强调中性、克制、避免敏感推断，并将长期特征与近期状态明确分层。 4. 重写成员最终画像生成逻辑，优先基于日/周/月摘要融合出长期特征、习惯模式、长期回复偏好、近期状态等信息，再用 AI 对分层摘要做最终整理，避免仅依赖近 30 天消息得出偏短期结论。 5. 保留并增强长期画像融合逻辑，通过打分、衰减和重复证据累积，使长期特征随着时间逐步稳定，而不会被单次刷新完全覆盖。 6. 在消息存储层补充成员按时间增量获取、按活跃日期统计、按天取消息等查询方法，为后续分层摘要生成提供数据支撑。 7. 扩展 member_context 插件配置，增加日级摘要消息上限、日摘要最小消息数、单次回填的日摘要数量上限、最终画像使用的日/周/月摘要数量等参数，便于在准确性和系统负载之间做平衡。 8. 后台成员摘要详情页新增长期沟通倾向、长期特征、习惯模式、长期回复偏好、近期状态、历史样本数、分层摘要数量等展示字段，方便观察画像沉淀程度。 9. 优化后台查看成员摘要接口逻辑：首次打开如果还没有摘要，不再同步阻塞生成，而是返回未就绪状态，配合后台手动异步刷新，降低页面卡顿和接口阻塞风险。 10. 增强刷新日志，单成员和群级刷新会输出当前刷新模式以及日/周/月摘要数量，便于排查画像构建进度。 11. 调整当前日、当前周、当前月摘要的重算逻辑，确保新增日摘要写入后，本周和本月摘要不会长期停留在旧版本。 本次提交后，成员画像能力从“基于近期样本的单层摘要”升级为“基于时间沉淀的分层画像管线”，为后续把画像稳定接入 AI 自动回复上下文打下基础，同时尽量保持现有群权限控制和后台异步刷新方式不变。
This commit is contained in:
liuwei
+367
-238
@@ -1,6 +1,5 @@
|
||||
# -*- coding: utf-8 -*-
|
||||
import json
|
||||
import math
|
||||
import re
|
||||
from collections import Counter
|
||||
from datetime import datetime
|
||||
@@ -12,7 +11,10 @@ from loguru import logger
|
||||
from db.connection import DBConnectionManager
|
||||
from db.contacts_db import ContactsDBOperator
|
||||
from db.member_context_db import MemberContextDBOperator
|
||||
from db.member_digest_db import MemberDigestDBOperator
|
||||
from db.message_storage import MessageStorageDB
|
||||
from plugins.member_context.digest_service import MemberDigestService
|
||||
from plugins.member_context.prompt_builder import MemberContextPromptBuilder
|
||||
from utils.robot_cmd.robot_command import Feature, GroupBotManager, PermissionStatus
|
||||
|
||||
|
||||
@@ -33,66 +35,122 @@ class MemberContextService:
|
||||
self.contacts_db = ContactsDBOperator(self.db_manager)
|
||||
self.message_db = MessageStorageDB(self.db_manager)
|
||||
self.member_context_db = MemberContextDBOperator(self.db_manager)
|
||||
self.member_digest_db = MemberDigestDBOperator(self.db_manager)
|
||||
self.digest_service = MemberDigestService(
|
||||
self.contacts_db, self.message_db, self.member_digest_db, plugin_config or {}
|
||||
)
|
||||
self.LOG = logger
|
||||
self.plugin_config = plugin_config or {}
|
||||
|
||||
api_config = self.plugin_config.get("api", {})
|
||||
profile_config = self.plugin_config.get("profile", {})
|
||||
schedule_config = self.plugin_config.get("schedule", {})
|
||||
|
||||
self.ai_enabled = bool(api_config.get("enabled", False))
|
||||
self.ai_enabled = bool(api_config.get("enable", api_config.get("enabled", False)))
|
||||
self.ai_base_url = (api_config.get("base_url") or "").rstrip("/")
|
||||
self.ai_api_key = api_config.get("api_key", "")
|
||||
self.ai_endpoint = str(api_config.get("endpoint", "completion-messages")).lstrip("/")
|
||||
self.ai_timeout = int(api_config.get("request_timeout", 60))
|
||||
|
||||
self.sample_days = int(profile_config.get("sample_days", 30))
|
||||
self.ai_sample_limit = int(profile_config.get("sample_message_limit", 80))
|
||||
self.refresh_limit_per_member = int(profile_config.get("refresh_limit_per_member", 200))
|
||||
self.ai_min_member_messages = int(profile_config.get("ai_min_member_messages", 12))
|
||||
self.active_member_hours = int(profile_config.get("active_member_hours", 72))
|
||||
self.min_member_messages = int(profile_config.get("min_member_messages", 3))
|
||||
self.max_members_per_group_per_run = int(profile_config.get("max_members_per_group_per_run", 30))
|
||||
self.stale_hours = int(profile_config.get("stale_hours", 24))
|
||||
schedule_config = self.plugin_config.get("schedule", {})
|
||||
self.stable_decay = float(profile_config.get("stable_decay", 0.96))
|
||||
self.stable_max_items = int(profile_config.get("stable_max_items", 6))
|
||||
self.stable_min_score = float(profile_config.get("stable_min_score", 0.9))
|
||||
self.stable_ready_days = int(profile_config.get("stable_ready_days", 180))
|
||||
self.only_recent_active_groups = bool(schedule_config.get("only_recent_active_groups", False))
|
||||
self.active_hours = int(schedule_config.get("active_hours", 72))
|
||||
self.min_group_messages = int(schedule_config.get("min_group_messages", 20))
|
||||
|
||||
def build_member_context(self, chatroom_id: str, wxid: str, days: Optional[int] = None,
|
||||
limit: Optional[int] = None) -> Dict:
|
||||
limit: Optional[int] = None, force_digest_rebuild: bool = False) -> Dict:
|
||||
days = days or self.sample_days
|
||||
limit = limit or self.refresh_limit_per_member
|
||||
|
||||
existing_context = self.member_context_db.get_member_context(chatroom_id, wxid)
|
||||
member = self.contacts_db.get_chatroom_member_info(chatroom_id, wxid) or {}
|
||||
messages = self.message_db.get_member_recent_messages(chatroom_id, wxid, days=days, limit=limit)
|
||||
recent_messages = self.message_db.get_member_recent_messages(chatroom_id, wxid, days=min(days, 7), limit=100)
|
||||
|
||||
display_name = member.get("display_name") or member.get("nick_name") or wxid
|
||||
activity_level = self._calc_activity_level(len(messages), days)
|
||||
message_pattern = self._build_message_pattern(messages)
|
||||
response_style_hint = self._build_response_style_hint(messages)
|
||||
topics = self._extract_keywords(messages, limit=5)
|
||||
recent_focus = self._extract_keywords(recent_messages, limit=4)
|
||||
confidence = self._calc_confidence(len(messages))
|
||||
|
||||
digest_snapshot = self.digest_service.ensure_member_digest_pipeline(
|
||||
chatroom_id, wxid, force=force_digest_rebuild
|
||||
)
|
||||
daily_digests = digest_snapshot.get("daily_digests", [])
|
||||
weekly_digests = digest_snapshot.get("weekly_digests", [])
|
||||
monthly_digests = digest_snapshot.get("monthly_digests", [])
|
||||
|
||||
recent_messages = self.message_db.get_member_recent_messages(chatroom_id, wxid, days=min(days, 7), limit=120)
|
||||
monthly_structured = [item.get("structured", {}) or {} for item in monthly_digests]
|
||||
weekly_structured = [item.get("structured", {}) or {} for item in weekly_digests]
|
||||
daily_structured = [item.get("structured", {}) or {} for item in daily_digests]
|
||||
|
||||
observation_days = self._calc_observation_days(daily_digests)
|
||||
activity_level = self._calc_activity_level(len(recent_messages), max(min(days, 7), 1))
|
||||
context = {
|
||||
"chatroom_id": chatroom_id,
|
||||
"wxid": wxid,
|
||||
"display_name": display_name,
|
||||
"activity_level": activity_level,
|
||||
"message_pattern": message_pattern,
|
||||
"interaction_style": self._build_interaction_style(messages),
|
||||
"response_style_hint": response_style_hint,
|
||||
"topics_of_interest": topics,
|
||||
"recent_focus": recent_focus,
|
||||
"summary_text": self._build_summary_text(activity_level, message_pattern, response_style_hint, topics, recent_focus),
|
||||
"confidence": confidence,
|
||||
"source_message_count": len(messages),
|
||||
"message_pattern": self._best_text(
|
||||
daily_structured, ["message_pattern"], default=self._build_message_pattern(recent_messages)
|
||||
),
|
||||
"interaction_style": self._best_text(
|
||||
daily_structured, ["interaction_style"], default=self._build_interaction_style(recent_messages)
|
||||
),
|
||||
"response_style_hint": self._build_response_style_hint_from_digests(
|
||||
daily_structured, weekly_structured, monthly_structured
|
||||
),
|
||||
"topics_of_interest": self._extract_scored_items(
|
||||
monthly_structured + weekly_structured, ["long_term_topics", "stable_topics", "topics"], limit=5
|
||||
),
|
||||
"recent_focus": self._extract_scored_items(daily_structured, ["topics"], limit=4),
|
||||
"summary_text": "",
|
||||
"confidence": self._calc_digest_confidence(monthly_digests, weekly_digests, daily_digests),
|
||||
"source_message_count": len(recent_messages),
|
||||
"source_days": days,
|
||||
"last_profiled_at": datetime.now().strftime("%Y-%m-%d %H:%M:%S"),
|
||||
"meta": self._build_meta(messages, recent_messages),
|
||||
"meta": {
|
||||
"stable_traits": self._extract_scored_items(
|
||||
monthly_structured + weekly_structured, ["stable_traits", "engagement_traits"], limit=self.stable_max_items
|
||||
),
|
||||
"habit_patterns": self._extract_scored_items(
|
||||
monthly_structured + weekly_structured + daily_structured,
|
||||
["habit_patterns", "habit_signals"], limit=self.stable_max_items
|
||||
),
|
||||
"long_term_reply_preferences": self._extract_scored_items(
|
||||
monthly_structured + weekly_structured, ["long_term_reply_preferences", "reply_preferences"], limit=4
|
||||
),
|
||||
"recent_state": self._extract_scored_items(
|
||||
weekly_structured + daily_structured, ["recent_state", "phase_state", "topics"], limit=4
|
||||
),
|
||||
"temperament_tendency": self._best_text(
|
||||
monthly_structured + weekly_structured + daily_structured,
|
||||
["temperament_tendency", "temperament_signal"], default=""
|
||||
),
|
||||
"engagement_traits": self._extract_scored_items(
|
||||
daily_structured + weekly_structured, ["engagement_traits", "stable_traits"], limit=4
|
||||
),
|
||||
"reply_taboos": self._extract_scored_items(daily_structured, ["reply_taboos"], limit=3),
|
||||
"observation_days": observation_days,
|
||||
"stable_ready": observation_days >= self.stable_ready_days,
|
||||
"profile_iterations": int(((existing_context or {}).get("meta", {}) or {}).get("profile_iterations", 0)) + 1,
|
||||
"history_message_count": self._sum_digest_source_count(daily_digests),
|
||||
"digest_daily_count": len(daily_digests),
|
||||
"digest_weekly_count": len(weekly_digests),
|
||||
"digest_monthly_count": len(monthly_digests),
|
||||
"last_daily_digest_at": daily_digests[0].get("last_generated_at") if daily_digests else "",
|
||||
"last_weekly_digest_at": weekly_digests[0].get("last_generated_at") if weekly_digests else "",
|
||||
"last_monthly_digest_at": monthly_digests[0].get("last_generated_at") if monthly_digests else "",
|
||||
"refresh_mode": self._build_refresh_mode(existing_context, digest_snapshot),
|
||||
},
|
||||
}
|
||||
|
||||
ai_context = self._generate_ai_context(chatroom_id, wxid, display_name, context, messages)
|
||||
ai_context = self._generate_ai_context_from_digests(
|
||||
chatroom_id, wxid, display_name, monthly_digests, weekly_digests, daily_digests
|
||||
)
|
||||
if ai_context:
|
||||
context.update({
|
||||
"activity_level": ai_context.get("activity_level") or context["activity_level"],
|
||||
@@ -105,6 +163,9 @@ class MemberContextService:
|
||||
"confidence": ai_context.get("confidence", context["confidence"]),
|
||||
})
|
||||
context["meta"].update(ai_context.get("meta", {}))
|
||||
|
||||
context = self._merge_with_existing_context(existing_context, context)
|
||||
context["summary_text"] = context.get("summary_text") or self._build_summary_text_from_context(context)
|
||||
return context
|
||||
|
||||
def refresh_member_context(self, chatroom_id: str, wxid: str, days: Optional[int] = None,
|
||||
@@ -117,6 +178,10 @@ class MemberContextService:
|
||||
self.LOG.info(
|
||||
f"[成员交互摘要] 单个成员刷新完成: group={chatroom_id}, wxid={wxid}, "
|
||||
f"display_name={context.get('display_name', wxid)}, messages={context.get('source_message_count', 0)}, "
|
||||
f"mode={context.get('meta', {}).get('refresh_mode', '')}, "
|
||||
f"digests={context.get('meta', {}).get('digest_daily_count', 0)}/"
|
||||
f"{context.get('meta', {}).get('digest_weekly_count', 0)}/"
|
||||
f"{context.get('meta', {}).get('digest_monthly_count', 0)}, "
|
||||
f"ai={'yes' if context.get('meta', {}).get('ai_provider') else 'no'}"
|
||||
)
|
||||
return context
|
||||
@@ -167,7 +232,7 @@ class MemberContextService:
|
||||
)
|
||||
continue
|
||||
context = self.build_member_context(chatroom_id, wxid, days=days, limit=limit_per_member)
|
||||
if context["source_message_count"] <= 0:
|
||||
if context["source_message_count"] <= 0 and context.get("meta", {}).get("digest_daily_count", 0) <= 0:
|
||||
skipped += 1
|
||||
self.LOG.debug(
|
||||
f"[成员交互摘要] 跳过成员(样本不足): group={chatroom_id}, "
|
||||
@@ -181,6 +246,10 @@ class MemberContextService:
|
||||
f"wxid={wxid}, display_name={context.get('display_name', wxid)}, "
|
||||
f"messages={context.get('source_message_count', 0)}, "
|
||||
f"activity={context.get('activity_level', '')}, "
|
||||
f"mode={context.get('meta', {}).get('refresh_mode', '')}, "
|
||||
f"digests={context.get('meta', {}).get('digest_daily_count', 0)}/"
|
||||
f"{context.get('meta', {}).get('digest_weekly_count', 0)}/"
|
||||
f"{context.get('meta', {}).get('digest_monthly_count', 0)}, "
|
||||
f"ai={'yes' if context.get('meta', {}).get('ai_provider') else 'no'}"
|
||||
)
|
||||
|
||||
@@ -238,7 +307,10 @@ class MemberContextService:
|
||||
f"skipped={result.get('skipped', 0)}, active_candidates={result.get('active_candidates', 0)}"
|
||||
)
|
||||
|
||||
self.LOG.info(f"成员交互摘要刷新完成: 启用活跃群={group_count}, 成员={member_count}, 跳过={skipped}, 未启用群={disabled}, 非活跃群={inactive}")
|
||||
self.LOG.info(
|
||||
f"成员交互摘要刷新完成: 启用活跃群={group_count}, 成员={member_count}, 跳过={skipped}, "
|
||||
f"未启用群={disabled}, 非活跃群={inactive}"
|
||||
)
|
||||
return {"groups": group_count, "members": member_count, "skipped": skipped, "disabled_groups": disabled, "inactive_groups": inactive}
|
||||
|
||||
def is_group_enabled(self, chatroom_id: str) -> bool:
|
||||
@@ -247,129 +319,6 @@ class MemberContextService:
|
||||
return True
|
||||
return GroupBotManager.get_group_permission(chatroom_id, feature) == PermissionStatus.ENABLED
|
||||
|
||||
def _calc_activity_level(self, message_count: int, days: int) -> str:
|
||||
daily_avg = message_count / max(days, 1)
|
||||
if message_count >= 80 or daily_avg >= 3:
|
||||
return "高活跃"
|
||||
if message_count >= 25 or daily_avg >= 1:
|
||||
return "中活跃"
|
||||
if message_count > 0:
|
||||
return "低活跃"
|
||||
return "观察中"
|
||||
|
||||
def _build_message_pattern(self, messages: List[Dict]) -> str:
|
||||
if not messages:
|
||||
return "样本较少，暂不做明显模式判断"
|
||||
|
||||
contents = [m.get("content", "") for m in messages if m.get("content")]
|
||||
if not contents:
|
||||
return "样本较少，暂不做明显模式判断"
|
||||
|
||||
avg_len = sum(len(c) for c in contents) / len(contents)
|
||||
question_ratio = sum(1 for c in contents if "?" in c or "？" in c) / len(contents)
|
||||
link_ratio = sum(1 for c in contents if "http://" in c or "https://" in c) / len(contents)
|
||||
|
||||
traits = []
|
||||
if avg_len <= 12:
|
||||
traits.append("短句居多")
|
||||
elif avg_len >= 35:
|
||||
traits.append("表达较完整")
|
||||
else:
|
||||
traits.append("表达中等长度")
|
||||
|
||||
if question_ratio >= 0.35:
|
||||
traits.append("问题导向明显")
|
||||
elif question_ratio >= 0.15:
|
||||
traits.append("偶尔连续追问")
|
||||
|
||||
if link_ratio >= 0.15:
|
||||
traits.append("常分享链接或资料")
|
||||
|
||||
if not traits:
|
||||
traits.append("发言较平稳")
|
||||
return "，".join(traits)
|
||||
|
||||
def _build_response_style_hint(self, messages: List[Dict]) -> str:
|
||||
if not messages:
|
||||
return "样本不足时保持中性、简洁、避免过度熟络"
|
||||
|
||||
contents = [m.get("content", "") for m in messages if m.get("content")]
|
||||
avg_len = sum(len(c) for c in contents) / max(len(contents), 1)
|
||||
question_ratio = sum(1 for c in contents if "?" in c or "？" in c) / max(len(contents), 1)
|
||||
|
||||
if question_ratio >= 0.35:
|
||||
return "优先给明确结论，再补充步骤或依据，避免空泛回应"
|
||||
if avg_len <= 12:
|
||||
return "回复尽量简洁直接，先回答核心点，减少铺垫"
|
||||
if avg_len >= 35:
|
||||
return "可以给稍完整的解释，但保持结构清楚，避免冗长"
|
||||
return "保持自然口语化，结论和解释尽量平衡"
|
||||
|
||||
def _build_interaction_style(self, messages: List[Dict]) -> str:
|
||||
if not messages:
|
||||
return "互动样本较少"
|
||||
contents = [m.get("content", "") for m in messages if m.get("content")]
|
||||
question_ratio = sum(1 for c in contents if "?" in c or "？" in c) / max(len(contents), 1)
|
||||
emoji_ratio = sum(1 for c in contents if re.search(r"[\U0001F300-\U0001FAFF\u2600-\u27BF]", c)) / max(len(contents), 1)
|
||||
mention_ratio = sum(1 for c in contents if "@" in c) / max(len(contents), 1)
|
||||
|
||||
parts = []
|
||||
if question_ratio >= 0.3:
|
||||
parts.append("偏提问推进")
|
||||
if emoji_ratio >= 0.15:
|
||||
parts.append("表情互动感较强")
|
||||
if mention_ratio >= 0.1:
|
||||
parts.append("会主动点名互动")
|
||||
if not parts:
|
||||
parts.append("自然跟随式互动")
|
||||
return "，".join(parts)
|
||||
|
||||
def _extract_keywords(self, messages: List[Dict], limit: int = 5) -> List[str]:
|
||||
counter = Counter()
|
||||
for message in messages:
|
||||
content = message.get("content", "")
|
||||
for token in self._tokenize(content):
|
||||
if token in self.STOPWORDS:
|
||||
continue
|
||||
counter[token] += 1
|
||||
return [word for word, _ in counter.most_common(limit)]
|
||||
|
||||
def _tokenize(self, text: str) -> List[str]:
|
||||
chinese_words = re.findall(r"[\u4e00-\u9fff]{2,6}", text)
|
||||
english_words = re.findall(r"[A-Za-z][A-Za-z0-9_-]{2,20}", text)
|
||||
return chinese_words + [word.lower() for word in english_words]
|
||||
|
||||
def _calc_confidence(self, message_count: int) -> float:
|
||||
return round(min(0.95, math.log(message_count + 1, 10)), 2) if message_count > 0 else 0.1
|
||||
|
||||
def _build_summary_text(self, activity_level: str, message_pattern: str,
|
||||
response_style_hint: str, topics: List[str], recent_focus: List[str]) -> str:
|
||||
parts = [
|
||||
f"近期互动强度：{activity_level}",
|
||||
f"表达特征：{message_pattern}",
|
||||
f"回复建议：{response_style_hint}",
|
||||
]
|
||||
if topics:
|
||||
parts.append(f"长期关注：{'、'.join(topics)}")
|
||||
if recent_focus:
|
||||
parts.append(f"近期话题：{'、'.join(recent_focus)}")
|
||||
return "；".join(parts)
|
||||
|
||||
def _build_meta(self, messages: List[Dict], recent_messages: List[Dict]) -> Dict:
|
||||
latest_time = None
|
||||
if recent_messages:
|
||||
latest = recent_messages[-1].get("timestamp")
|
||||
if isinstance(latest, datetime):
|
||||
latest_time = latest.strftime("%Y-%m-%d %H:%M:%S")
|
||||
elif latest:
|
||||
latest_time = str(latest)
|
||||
|
||||
return {
|
||||
"message_count_30d": len(messages),
|
||||
"message_count_7d": len(recent_messages),
|
||||
"latest_message_time": latest_time,
|
||||
}
|
||||
|
||||
def _get_recent_active_members(self, chatroom_id: str) -> List[Dict]:
|
||||
sql = """
|
||||
SELECT
|
||||
@@ -425,7 +374,10 @@ class MemberContextService:
|
||||
try:
|
||||
return datetime.strptime(str(value), "%Y-%m-%d %H:%M:%S")
|
||||
except Exception:
|
||||
return None
|
||||
try:
|
||||
return datetime.strptime(str(value)[:10], "%Y-%m-%d")
|
||||
except Exception:
|
||||
return None
|
||||
|
||||
def _get_recent_active_chatrooms(self) -> set:
|
||||
sql = """
|
||||
@@ -439,14 +391,17 @@ class MemberContextService:
|
||||
rows = self.message_db.execute_query(sql, ("%@chatroom", self.active_hours, self.min_group_messages)) or []
|
||||
return {row.get("group_id") for row in rows if row.get("group_id")}
|
||||
|
||||
def _generate_ai_context(self, chatroom_id: str, wxid: str, display_name: str,
|
||||
base_context: Dict, messages: List[Dict]) -> Optional[Dict]:
|
||||
def _generate_ai_context_from_digests(self, chatroom_id: str, wxid: str, display_name: str,
|
||||
monthly_digests: List[Dict], weekly_digests: List[Dict],
|
||||
daily_digests: List[Dict]) -> Optional[Dict]:
|
||||
if not self.ai_enabled or not self.ai_base_url or not self.ai_api_key:
|
||||
return None
|
||||
if len(messages) < self.ai_min_member_messages:
|
||||
if len(daily_digests) < 2 and len(weekly_digests) < 1 and len(monthly_digests) < 1:
|
||||
return None
|
||||
|
||||
prompt = self._build_ai_prompt(chatroom_id, wxid, display_name, base_context, messages[-self.ai_sample_limit:])
|
||||
prompt = MemberContextPromptBuilder.build_final_context_prompt(
|
||||
chatroom_id, wxid, display_name, monthly_digests, weekly_digests, daily_digests
|
||||
)
|
||||
headers = {
|
||||
"Authorization": f"Bearer {self.ai_api_key}",
|
||||
"Content-Type": "application/json",
|
||||
@@ -454,88 +409,38 @@ class MemberContextService:
|
||||
payload = {
|
||||
"inputs": {"query": prompt},
|
||||
"response_mode": "blocking",
|
||||
"user": f"member-context:{chatroom_id}:{wxid}",
|
||||
"user": f"member-context-final:{chatroom_id}:{wxid}",
|
||||
}
|
||||
url = f"{self.ai_base_url}/{self.ai_endpoint}"
|
||||
try:
|
||||
self.LOG.info(
|
||||
f"[成员交互摘要][AI] 发起最终画像请求: group={chatroom_id}, wxid={wxid}, "
|
||||
f"monthly={len(monthly_digests)}, weekly={len(weekly_digests)}, daily={len(daily_digests)}"
|
||||
)
|
||||
response = requests.post(url, headers=headers, json=payload, timeout=self.ai_timeout)
|
||||
response.raise_for_status()
|
||||
response_data = response.json()
|
||||
parsed = self._parse_ai_answer(response_data.get("answer", ""))
|
||||
data = response.json()
|
||||
parsed = self._parse_ai_answer(data.get("answer", ""))
|
||||
if not parsed:
|
||||
self.LOG.warning(
|
||||
f"[成员交互摘要][AI] 最终画像JSON解析失败: group={chatroom_id}, wxid={wxid}, "
|
||||
f"answer_preview={(data.get('answer', '') or '')[:200]}"
|
||||
)
|
||||
return None
|
||||
usage = (response_data.get("metadata") or {}).get("usage", {}) or {}
|
||||
parsed["meta"] = {
|
||||
usage = (data.get("metadata") or {}).get("usage", {}) or {}
|
||||
parsed_meta = parsed.get("meta", {}) or {}
|
||||
parsed_meta.update({
|
||||
"ai_provider": "dify",
|
||||
"ai_mode": "completion",
|
||||
"ai_tokens": usage.get("total_tokens"),
|
||||
"ai_latency": usage.get("latency"),
|
||||
}
|
||||
})
|
||||
parsed["meta"] = parsed_meta
|
||||
return parsed
|
||||
except Exception as e:
|
||||
self.LOG.warning(f"成员交互摘要 AI 生成失败，回退到本地摘要: chatroom={chatroom_id}, wxid={wxid}, error={e}")
|
||||
self.LOG.warning(f"成员交互摘要最终画像 AI 生成失败，回退到本地融合: chatroom={chatroom_id}, wxid={wxid}, error={e}")
|
||||
return None
|
||||
|
||||
def _build_ai_prompt(self, chatroom_id: str, wxid: str, display_name: str,
|
||||
base_context: Dict, messages: List[Dict]) -> str:
|
||||
message_lines = []
|
||||
for msg in messages[-40:]:
|
||||
ts = msg.get("timestamp")
|
||||
if isinstance(ts, datetime):
|
||||
ts = ts.strftime("%m-%d %H:%M")
|
||||
content = (msg.get("content") or "").replace("\n", " ").strip()
|
||||
content = content[:160]
|
||||
if content:
|
||||
message_lines.append(f"[{ts}] {content}")
|
||||
|
||||
topics = "、".join(base_context.get("topics_of_interest", [])) or "无明显长期话题"
|
||||
recent_focus = "、".join(base_context.get("recent_focus", [])) or "无明显近期话题"
|
||||
|
||||
return (
|
||||
"你是一个微信群运营后台的成员交互摘要提取器。\n"
|
||||
"你的任务不是做人设分析，也不是性格判断，而是基于公开聊天记录，提取对后续回复策略有帮助的“交互特征摘要”。\n"
|
||||
"你只能依据给定聊天样本输出保守结论，不能脑补，不能做敏感推断，不能写负面标签，不能输出隐私猜测。\n"
|
||||
"请根据以下成员近30天公开发言，输出一个严格 JSON 对象，不要 markdown，不要解释，不要代码块。\n"
|
||||
"JSON schema:\n"
|
||||
"{"
|
||||
"\"activity_level\":\"高活跃|中活跃|低活跃|观察中\","
|
||||
"\"message_pattern\":\"一句中文，描述表达特点\","
|
||||
"\"interaction_style\":\"一句中文，描述他在群里如何与人互动\","
|
||||
"\"response_style_hint\":\"一句中文，描述适合怎样回应\","
|
||||
"\"topics_of_interest\":[\"主题1\",\"主题2\"],"
|
||||
"\"recent_focus\":[\"近期主题1\",\"近期主题2\"],"
|
||||
"\"summary_text\":\"一段不超过120字的后台交互摘要\","
|
||||
"\"confidence\":0.0,"
|
||||
"\"engagement_traits\":[\"特征1\",\"特征2\"],"
|
||||
"\"reply_taboos\":[\"避坑1\",\"避坑2\"]"
|
||||
"}\n"
|
||||
"要求：\n"
|
||||
"1. 只总结群内公开行为特征，不要输出性格诊断、负面标签或敏感结论。\n"
|
||||
"2. topics_of_interest 表示相对稳定的话题偏好，最多5个；recent_focus 表示近期频繁提及的话题，最多4个。\n"
|
||||
"3. message_pattern 只能描述可观察到的表达方式，例如：短句居多、问题导向、爱发链接、解释较完整、常接梗互动。\n"
|
||||
"4. interaction_style 要描述他在群里的参与方式，例如：偏围观后插话、喜欢接梗、会连续追问、偏一对一回应。\n"
|
||||
"5. response_style_hint 只能写对回复策略有帮助的建议，例如：先给结论再补步骤、保持简洁直接、可以适度接梗；不要写成评价语。\n"
|
||||
"6. engagement_traits 最多4个，写成中性的短标签，例如：节奏快、爱追问细节、接梗自然、偏结果导向。\n"
|
||||
"7. reply_taboos 最多3个，只写回复时应避免的方式，例如：避免长篇铺垫、避免过度说教、避免太官方。\n"
|
||||
"8. summary_text 要像后台备注，客观、中性、克制，不要让人一眼看出是在给用户贴标签。\n"
|
||||
"9. confidence 取值 0 到 1；如果样本较少或不稳定，必须降低 confidence。\n"
|
||||
"10. 如果证据不足，宁可输出更弱、更泛化的结论，也不要瞎猜。\n\n"
|
||||
"下面是正反例参考。\n"
|
||||
"坏例子：这个人情绪化、爱抬杠、虚荣、玻璃心。\n"
|
||||
"好例子：常用短句直接表达观点；遇到问题时更适合先给明确结论，再补充解释。\n\n"
|
||||
f"成员标识: {display_name} ({wxid})\n"
|
||||
f"群ID: {chatroom_id}\n"
|
||||
f"样本消息数: {base_context.get('source_message_count', 0)}\n"
|
||||
f"本地活跃度估计: {base_context.get('activity_level', '')}\n"
|
||||
f"本地表达特征: {base_context.get('message_pattern', '')}\n"
|
||||
f"本地互动风格: {base_context.get('interaction_style', '')}\n"
|
||||
f"本地回复建议: {base_context.get('response_style_hint', '')}\n"
|
||||
f"本地长期关注: {topics}\n"
|
||||
f"本地近期话题: {recent_focus}\n"
|
||||
"最近消息样本:\n"
|
||||
+ "\n".join(message_lines)
|
||||
)
|
||||
|
||||
def _parse_ai_answer(self, answer: str) -> Optional[Dict]:
|
||||
if not answer:
|
||||
return None
|
||||
@@ -548,18 +453,10 @@ class MemberContextService:
|
||||
except Exception:
|
||||
return None
|
||||
|
||||
topics = data.get("topics_of_interest") or []
|
||||
recent_focus = data.get("recent_focus") or []
|
||||
engagement_traits = data.get("engagement_traits") or []
|
||||
reply_taboos = data.get("reply_taboos") or []
|
||||
if not isinstance(topics, list):
|
||||
topics = []
|
||||
if not isinstance(recent_focus, list):
|
||||
recent_focus = []
|
||||
if not isinstance(engagement_traits, list):
|
||||
engagement_traits = []
|
||||
if not isinstance(reply_taboos, list):
|
||||
reply_taboos = []
|
||||
def norm_list(value, limit):
|
||||
if not isinstance(value, list):
|
||||
return []
|
||||
return [str(item).strip() for item in value[:limit] if str(item).strip()]
|
||||
|
||||
try:
|
||||
confidence = float(data.get("confidence", 0))
|
||||
@@ -571,12 +468,244 @@ class MemberContextService:
|
||||
"message_pattern": str(data.get("message_pattern", "")).strip(),
|
||||
"interaction_style": str(data.get("interaction_style", "")).strip(),
|
||||
"response_style_hint": str(data.get("response_style_hint", "")).strip(),
|
||||
"topics_of_interest": [str(item).strip() for item in topics[:5] if str(item).strip()],
|
||||
"recent_focus": [str(item).strip() for item in recent_focus[:4] if str(item).strip()],
|
||||
"topics_of_interest": norm_list(data.get("topics_of_interest"), 5),
|
||||
"recent_focus": norm_list(data.get("recent_focus"), 4),
|
||||
"summary_text": str(data.get("summary_text", "")).strip(),
|
||||
"confidence": max(0.0, min(1.0, confidence)),
|
||||
"meta": {
|
||||
"engagement_traits": [str(item).strip() for item in engagement_traits[:4] if str(item).strip()],
|
||||
"reply_taboos": [str(item).strip() for item in reply_taboos[:3] if str(item).strip()],
|
||||
"stable_traits": norm_list(data.get("stable_traits"), self.stable_max_items),
|
||||
"habit_patterns": norm_list(data.get("habit_patterns"), self.stable_max_items),
|
||||
"long_term_reply_preferences": norm_list(data.get("long_term_reply_preferences"), 4),
|
||||
"recent_state": norm_list(data.get("recent_state"), 4),
|
||||
"temperament_tendency": str(data.get("temperament_tendency", "")).strip(),
|
||||
"engagement_traits": norm_list(data.get("engagement_traits"), 4),
|
||||
"reply_taboos": norm_list(data.get("reply_taboos"), 3),
|
||||
}
|
||||
}
|
||||
|
||||
def _merge_with_existing_context(self, existing_context: Optional[Dict], current_context: Dict) -> Dict:
|
||||
existing_context = existing_context or {}
|
||||
existing_meta = existing_context.get("meta", {}) or {}
|
||||
meta = current_context.get("meta", {}) or {}
|
||||
|
||||
observation_days = max(
|
||||
int(meta.get("observation_days", 0)),
|
||||
int(existing_meta.get("observation_days", 0)),
|
||||
)
|
||||
meta["observation_days"] = observation_days
|
||||
meta["stable_ready"] = observation_days >= self.stable_ready_days
|
||||
|
||||
merged_topic_scores = self._merge_scored_items(
|
||||
existing_meta.get("topic_scores", {}),
|
||||
current_context.get("topics_of_interest", []),
|
||||
current_context.get("confidence", 0),
|
||||
)
|
||||
merged_trait_scores = self._merge_scored_items(
|
||||
existing_meta.get("stable_trait_scores", {}),
|
||||
meta.get("stable_traits", []),
|
||||
current_context.get("confidence", 0),
|
||||
)
|
||||
merged_habit_scores = self._merge_scored_items(
|
||||
existing_meta.get("habit_pattern_scores", {}),
|
||||
meta.get("habit_patterns", []),
|
||||
current_context.get("confidence", 0),
|
||||
)
|
||||
merged_reply_pref_scores = self._merge_scored_items(
|
||||
existing_meta.get("long_term_reply_preference_scores", {}),
|
||||
meta.get("long_term_reply_preferences", []),
|
||||
current_context.get("confidence", 0),
|
||||
)
|
||||
merged_temperament_scores = self._merge_scored_items(
|
||||
existing_meta.get("temperament_tendency_scores", {}),
|
||||
[meta.get("temperament_tendency")] if meta.get("temperament_tendency") else [],
|
||||
current_context.get("confidence", 0) * 0.9,
|
||||
)
|
||||
|
||||
meta["topic_scores"] = merged_topic_scores
|
||||
meta["stable_trait_scores"] = merged_trait_scores
|
||||
meta["habit_pattern_scores"] = merged_habit_scores
|
||||
meta["long_term_reply_preference_scores"] = merged_reply_pref_scores
|
||||
meta["temperament_tendency_scores"] = merged_temperament_scores
|
||||
meta["stable_traits"] = self._top_scored_items(merged_trait_scores, limit=self.stable_max_items)
|
||||
meta["habit_patterns"] = self._top_scored_items(merged_habit_scores, limit=self.stable_max_items)
|
||||
meta["long_term_reply_preferences"] = self._top_scored_items(merged_reply_pref_scores, limit=4)
|
||||
temperament = self._top_scored_items(merged_temperament_scores, limit=1)
|
||||
meta["temperament_tendency"] = temperament[0] if temperament else meta.get("temperament_tendency", "")
|
||||
meta["engagement_traits"] = (meta.get("engagement_traits") or existing_meta.get("engagement_traits") or [])[:4]
|
||||
meta["reply_taboos"] = (meta.get("reply_taboos") or existing_meta.get("reply_taboos") or [])[:3]
|
||||
meta["recent_state"] = (meta.get("recent_state") or existing_meta.get("recent_state") or [])[:4]
|
||||
meta["profile_iterations"] = max(
|
||||
int(meta.get("profile_iterations", 0)),
|
||||
int(existing_meta.get("profile_iterations", 0)),
|
||||
)
|
||||
meta["history_message_count"] = max(
|
||||
int(meta.get("history_message_count", 0)),
|
||||
int(existing_meta.get("history_message_count", 0)),
|
||||
)
|
||||
|
||||
current_context["topics_of_interest"] = self._top_scored_items(merged_topic_scores, limit=5) or current_context.get("topics_of_interest", [])
|
||||
current_context["recent_focus"] = (current_context.get("recent_focus") or existing_context.get("recent_focus") or [])[:4]
|
||||
current_context["response_style_hint"] = current_context.get("response_style_hint") or existing_context.get("response_style_hint") or ""
|
||||
current_context["meta"] = meta
|
||||
return current_context
|
||||
|
||||
def _extract_scored_items(self, items: List[Dict], keys: List[str], limit: int) -> List[str]:
|
||||
scores = {}
|
||||
for index, item in enumerate(items):
|
||||
weight = max(0.5, 1.2 - index * 0.08)
|
||||
for key in keys:
|
||||
values = item.get(key, [])
|
||||
if not isinstance(values, list):
|
||||
continue
|
||||
for value in values:
|
||||
normalized = str(value).strip()
|
||||
if not normalized:
|
||||
continue
|
||||
scores[normalized] = scores.get(normalized, 0.0) + weight
|
||||
return [key for key, _ in sorted(scores.items(), key=lambda pair: pair[1], reverse=True)[:limit]]
|
||||
|
||||
def _best_text(self, items: List[Dict], keys: List[str], default: str = "") -> str:
|
||||
counter = Counter()
|
||||
for item in items:
|
||||
for key in keys:
|
||||
value = str(item.get(key, "")).strip()
|
||||
if value:
|
||||
counter[value] += 1
|
||||
if counter:
|
||||
return counter.most_common(1)[0][0]
|
||||
return default
|
||||
|
||||
def _build_response_style_hint_from_digests(self, daily_structured: List[Dict],
|
||||
weekly_structured: List[Dict],
|
||||
monthly_structured: List[Dict]) -> str:
|
||||
hint = self._best_text(daily_structured, ["response_style_hint"])
|
||||
if hint:
|
||||
return hint
|
||||
preferences = self._extract_scored_items(
|
||||
monthly_structured + weekly_structured,
|
||||
["long_term_reply_preferences", "reply_preferences"],
|
||||
limit=3,
|
||||
)
|
||||
if preferences:
|
||||
return "更适合：" + "、".join(preferences[:3])
|
||||
return "保持自然口语化，结论和解释尽量平衡"
|
||||
|
||||
def _calc_digest_confidence(self, monthly_digests: List[Dict], weekly_digests: List[Dict],
|
||||
daily_digests: List[Dict]) -> float:
|
||||
base = 0.25
|
||||
base += min(0.35, len(monthly_digests) * 0.08)
|
||||
base += min(0.2, len(weekly_digests) * 0.04)
|
||||
base += min(0.15, len(daily_digests) * 0.02)
|
||||
return round(min(0.95, base), 2)
|
||||
|
||||
def _calc_observation_days(self, daily_digests: List[Dict]) -> int:
|
||||
if not daily_digests:
|
||||
return 0
|
||||
end_dt = self._parse_datetime(daily_digests[0].get("period_end"))
|
||||
start_dt = self._parse_datetime(daily_digests[-1].get("period_start"))
|
||||
if not start_dt or not end_dt:
|
||||
return 0
|
||||
return max(0, (end_dt - start_dt).days)
|
||||
|
||||
@staticmethod
|
||||
def _sum_digest_source_count(daily_digests: List[Dict]) -> int:
|
||||
return sum(int(item.get("source_count", 0)) for item in daily_digests)
|
||||
|
||||
def _build_refresh_mode(self, existing_context: Optional[Dict], digest_snapshot: Dict) -> str:
|
||||
if not existing_context:
|
||||
return "bootstrap"
|
||||
if (digest_snapshot.get("stats", {}) or {}).get("built_monthly", 0) > 0:
|
||||
return "recalibration"
|
||||
return "incremental"
|
||||
|
||||
def _build_summary_text_from_context(self, context: Dict) -> str:
|
||||
meta = context.get("meta", {}) or {}
|
||||
parts = []
|
||||
if meta.get("temperament_tendency"):
|
||||
label = "长期沟通倾向" if meta.get("stable_ready") else "阶段性沟通倾向"
|
||||
parts.append(f"{label}：{meta.get('temperament_tendency')}")
|
||||
if meta.get("stable_traits"):
|
||||
parts.append(f"长期特征：{'、'.join(meta.get('stable_traits')[:3])}")
|
||||
if meta.get("habit_patterns"):
|
||||
parts.append(f"习惯模式：{'、'.join(meta.get('habit_patterns')[:3])}")
|
||||
if meta.get("recent_state"):
|
||||
parts.append(f"近期状态：{'、'.join(meta.get('recent_state')[:3])}")
|
||||
if context.get("response_style_hint"):
|
||||
parts.append(f"回复建议：{context.get('response_style_hint')}")
|
||||
return "；".join(parts[:5])
|
||||
|
||||
def _merge_scored_items(self, existing_scores: Dict, current_items: List[str], confidence: float) -> Dict[str, float]:
|
||||
merged = {}
|
||||
for key, value in (existing_scores or {}).items():
|
||||
try:
|
||||
score = float(value) * self.stable_decay
|
||||
except Exception:
|
||||
continue
|
||||
if score >= 0.2:
|
||||
merged[str(key).strip()] = round(score, 4)
|
||||
|
||||
boost = max(0.6, min(1.8, 0.8 + confidence))
|
||||
for item in current_items or []:
|
||||
normalized = str(item).strip()
|
||||
if not normalized:
|
||||
continue
|
||||
merged[normalized] = round(merged.get(normalized, 0.0) + boost, 4)
|
||||
return merged
|
||||
|
||||
def _top_scored_items(self, scores: Dict, limit: int) -> List[str]:
|
||||
ordered = sorted(
|
||||
((str(key).strip(), float(value)) for key, value in (scores or {}).items() if str(key).strip()),
|
||||
key=lambda item: item[1],
|
||||
reverse=True,
|
||||
)
|
||||
return [key for key, value in ordered if value >= self.stable_min_score][:limit]
|
||||
|
||||
def _calc_activity_level(self, message_count: int, days: int) -> str:
|
||||
daily_avg = message_count / max(days, 1)
|
||||
if message_count >= 80 or daily_avg >= 3:
|
||||
return "高活跃"
|
||||
if message_count >= 25 or daily_avg >= 1:
|
||||
return "中活跃"
|
||||
if message_count > 0:
|
||||
return "低活跃"
|
||||
return "观察中"
|
||||
|
||||
def _build_message_pattern(self, messages: List[Dict]) -> str:
|
||||
if not messages:
|
||||
return "样本较少，暂不做明显模式判断"
|
||||
contents = [m.get("content", "") for m in messages if m.get("content")]
|
||||
if not contents:
|
||||
return "样本较少，暂不做明显模式判断"
|
||||
avg_len = sum(len(c) for c in contents) / len(contents)
|
||||
question_ratio = sum(1 for c in contents if "?" in c or "？" in c) / len(contents)
|
||||
link_ratio = sum(1 for c in contents if "http://" in c or "https://" in c) / len(contents)
|
||||
traits = []
|
||||
if avg_len <= 12:
|
||||
traits.append("短句居多")
|
||||
elif avg_len >= 35:
|
||||
traits.append("表达较完整")
|
||||
else:
|
||||
traits.append("表达中等长度")
|
||||
if question_ratio >= 0.35:
|
||||
traits.append("问题导向明显")
|
||||
elif question_ratio >= 0.15:
|
||||
traits.append("偶尔连续追问")
|
||||
if link_ratio >= 0.15:
|
||||
traits.append("常分享链接或资料")
|
||||
return "，".join(traits or ["发言较平稳"])
|
||||
|
||||
def _build_interaction_style(self, messages: List[Dict]) -> str:
|
||||
if not messages:
|
||||
return "互动样本较少"
|
||||
contents = [m.get("content", "") for m in messages if m.get("content")]
|
||||
question_ratio = sum(1 for c in contents if "?" in c or "？" in c) / max(len(contents), 1)
|
||||
emoji_ratio = sum(1 for c in contents if re.search(r"[\U0001F300-\U0001FAFF\u2600-\u27BF]", c)) / max(len(contents), 1)
|
||||
mention_ratio = sum(1 for c in contents if "@" in c) / max(len(contents), 1)
|
||||
parts = []
|
||||
if question_ratio >= 0.3:
|
||||
parts.append("偏提问推进")
|
||||
if emoji_ratio >= 0.15:
|
||||
parts.append("表情互动感较强")
|
||||
if mention_ratio >= 0.1:
|
||||
parts.append("会主动点名互动")
|
||||
return "，".join(parts or ["自然跟随式互动"])
|
||||
|
||||
Reference in New Issue
Block a user