Watch
1
0
Fork
You've already forked arc
0
forked from hesabix/arc
arc/hesabixAPI/app/services/ai/ai_memory_feedback_service.py

91 lines
2.8 KiB
Python

"""یادگیری از بازخورد thumbs — به‌صورت آیتم حافظه (بی‌صدا)."""
from __future__ import annotations
import logging
from datetime import datetime
from typing import Dict, Optional, Tuple
from sqlalchemy.orm import Session
from adapters.db.models.ai_chat_message import AIChatMessage
from adapters.db.models.ai_chat_session import AIChatSession
from app.services.ai.ai_constants import MAX_MEMORY_FEEDBACK_UPDATES_PER_DAY
from app.services.ai.ai_ops_metrics import log_ai_event
logger = logging.getLogger(__name__)
_feedback_daily: Dict[Tuple[int, int], Tuple[str, int]] = {}
def _day_key() -> str:
return datetime.utcnow().strftime("%Y-%m-%d")
def _can_apply_feedback(user_id: int, business_id: int) -> bool:
key = (int(user_id), int(business_id))
day = _day_key()
entry = _feedback_daily.get(key)
if not entry or entry[0] != day:
return True
return entry[1] < MAX_MEMORY_FEEDBACK_UPDATES_PER_DAY
def _record_feedback_apply(user_id: int, business_id: int) -> None:
key = (int(user_id), int(business_id))
day = _day_key()
entry = _feedback_daily.get(key)
if not entry or entry[0] != day:
_feedback_daily[key] = (day, 1)
else:
_feedback_daily[key] = (day, entry[1] + 1)
def apply_feedback_to_memory(
db: Session,
message_id: int,
user_id: int,
rating: int,
comment: Optional[str] = None,
) -> bool:
msg = db.query(AIChatMessage).filter(AIChatMessage.id == message_id).first()
if not msg or msg.role != "assistant":
return False
session = db.query(AIChatSession).filter(AIChatSession.id == msg.session_id).first()
if not session or not session.business_id:
return False
business_id = int(session.business_id)
if not _can_apply_feedback(user_id, business_id):
return False
# فقط بازخورد منفی با نظر متنی ارزش حافظه دارد
if rating >= 0 or not (comment or "").strip():
return False
from app.services.ai.ai_memory_item_service import upsert_memory_item
note = f"اجتناب از این الگو: {comment.strip()[:160]}"
try:
upsert_memory_item(
db,
business_id,
user_id,
item_key=None,
category="constraint",
content=note,
source="feedback",
confidence="medium",
)
_record_feedback_apply(user_id, business_id)
log_ai_event(
"memory_feedback_applied",
business_id=business_id,
user_id=user_id,
session_id=session.id,
extra={"rating": rating, "message_id": message_id},
)
return True
except Exception as exc:
logger.warning("apply_feedback_to_memory failed: %s", exc)
return False