Files
life-echo/api/app/features/memory/enrichment.py
Kevin 41518bda11 聊天和回忆录证据检索都走 pgvector,去掉 Postgres FTS/content_tsv,新迁移删掉 content_tsv 列(部署要先 alembic upgrade)。
Embedding 端口增加 is_available(),聊天和回忆录日志用统一方式表示向量是否真能调用。

记忆整理(compaction)支持 Beat 定期扫用户;

事实抽取提示与 subject 归一化,减少同一人多种称呼;
2026-04-03 11:43:16 +08:00

295 lines
9.3 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
"""
Transcript ingest 之后的记忆富化:摘要、事实、时间线。
由 Celerysync与 MemoryService.ingestasync调用失败仅打日志不阻断主流程。
"""
from __future__ import annotations
from typing import Any
from sqlalchemy import select
from sqlalchemy.ext.asyncio import AsyncSession
from sqlalchemy.orm import Session
from app.core.logging import get_logger
from app.features.memory.extractor import (
extract_facts_from_transcript_async,
extract_facts_from_transcript_sync,
)
from app.features.memory.models import MemoryChunk, MemorySummary
from app.features.memory.repo import (
create_memory_fact,
create_memory_fact_sync,
create_memory_summary,
create_memory_summary_sync,
create_timeline_event,
create_timeline_event_sync,
delete_timeline_events_by_memory_source,
delete_timeline_events_by_memory_source_sync,
list_chunks_for_source_sync,
upsert_rolling_summary_sync,
)
from app.features.memory.summarizer import (
generate_rolling_summary_async,
generate_rolling_summary_sync,
generate_session_summary_async,
generate_session_summary_sync,
)
from app.features.memory.enrichment_pipeline import (
dedupe_key,
normalize_object_json,
normalize_subject,
)
from app.features.user.models import User
from app.features.memory.timeline import (
build_timeline_events_from_facts_async,
build_timeline_events_from_facts_sync,
)
logger = get_logger(__name__)
def _resolve_llm_sync() -> Any | None:
try:
from app.core.dependencies import get_llm_provider_fast
return get_llm_provider_fast().langchain_llm
except Exception as e:
logger.warning("memory enrichment 无法获取 LLM: {}", e)
return None
def enrich_memory_after_ingest_sync(
session: Session,
user_id: str,
source_id: str,
llm: Any | None = None,
) -> None:
from app.core.config import settings
if not settings.memory_enrichment_enabled:
return
if llm is None:
llm = _resolve_llm_sync()
if not llm:
return
narrator_name: str | None = None
u_row = session.get(User, user_id)
if u_row and (u_row.nickname or "").strip():
narrator_name = (u_row.nickname or "").strip()
chunks = list_chunks_for_source_sync(session, source_id)
if not chunks:
return
chunk_texts = [c.content for c in chunks]
chunk_ids = [c.id for c in chunks]
numbered = "\n\n".join(
f"[chunk_id={cid}]\n{txt}" for cid, txt in zip(chunk_ids, chunk_texts)
)
session_summary_text = generate_session_summary_sync(llm, chunk_texts)
if session_summary_text:
create_memory_summary_sync(
session,
user_id=user_id,
summary_type="session",
content=session_summary_text,
source_chunk_ids=chunk_ids,
)
existing_rolling = (
session.execute(
select(MemorySummary)
.where(
MemorySummary.user_id == user_id,
MemorySummary.summary_type == "rolling",
)
.order_by(MemorySummary.updated_at.desc())
.limit(1)
)
.unique()
.scalar_one_or_none()
)
existing_text = existing_rolling.content if existing_rolling else None
rolling_text = generate_rolling_summary_sync(llm, existing_text, chunk_texts)
if rolling_text:
upsert_rolling_summary_sync(
session,
user_id=user_id,
content=rolling_text,
source_chunk_ids=chunk_ids,
)
raw_facts = extract_facts_from_transcript_sync(
llm, numbered, narrator_name=narrator_name
)
seen: set[tuple] = set()
inserted: list[dict] = []
for f in raw_facts:
key = dedupe_key(f, narrator_name=narrator_name)
if key in seen:
continue
seen.add(key)
scid = f.get("source_chunk_id")
if scid and scid not in chunk_ids:
scid = chunk_ids[0] if chunk_ids else None
row = create_memory_fact_sync(
session,
user_id=user_id,
fact_type=f.get("fact_type") or "event",
subject=normalize_subject(f.get("subject"), narrator_name),
predicate=f.get("predicate"),
object_json=normalize_object_json(f.get("object_json")),
confidence=float(f.get("confidence") or 0.75),
source_chunk_id=scid,
status="confirmed",
)
inserted.append(
{
"id": row.id,
"fact_type": row.fact_type,
"subject": row.subject,
"predicate": row.predicate,
"object_json": row.object_json,
}
)
if inserted:
delete_timeline_events_by_memory_source_sync(
session, user_id=user_id, memory_source_id=source_id
)
events = build_timeline_events_from_facts_sync(llm, inserted)
for ev in events:
create_timeline_event_sync(
session,
user_id=user_id,
event_year=ev.get("event_year"),
event_date=ev.get("event_date"),
title=ev["title"],
description=ev.get("description"),
source_fact_ids=ev.get("source_fact_ids") or None,
memory_source_id=source_id,
)
async def enrich_memory_after_ingest_async(
db: AsyncSession,
user_id: str,
source_id: str,
llm: Any | None = None,
) -> None:
from app.core.config import settings
if not settings.memory_enrichment_enabled:
return
if llm is None:
llm = _resolve_llm_sync()
if not llm:
return
narrator_name: str | None = None
u_row = await db.get(User, user_id)
if u_row and (u_row.nickname or "").strip():
narrator_name = (u_row.nickname or "").strip()
stmt = (
select(MemoryChunk)
.where(MemoryChunk.source_id == source_id)
.order_by(MemoryChunk.chunk_index.asc())
)
result = await db.execute(stmt)
chunks = list(result.unique().scalars().all())
if not chunks:
return
chunk_texts = [c.content for c in chunks]
chunk_ids = [c.id for c in chunks]
numbered = "\n\n".join(
f"[chunk_id={cid}]\n{txt}" for cid, txt in zip(chunk_ids, chunk_texts)
)
session_summary_text = await generate_session_summary_async(llm, chunk_texts)
if session_summary_text:
await create_memory_summary(
db,
user_id=user_id,
summary_type="session",
content=session_summary_text,
source_chunk_ids=chunk_ids,
)
roll_stmt = (
select(MemorySummary)
.where(
MemorySummary.user_id == user_id,
MemorySummary.summary_type == "rolling",
)
.order_by(MemorySummary.updated_at.desc())
.limit(1)
)
r_result = await db.execute(roll_stmt)
existing_row = r_result.unique().scalar_one_or_none()
existing_text = existing_row.content if existing_row else None
rolling_text = await generate_rolling_summary_async(llm, existing_text, chunk_texts)
if rolling_text:
if existing_row:
existing_row.content = rolling_text
existing_row.source_chunk_ids = chunk_ids
else:
await create_memory_summary(
db,
user_id=user_id,
summary_type="rolling",
content=rolling_text,
source_chunk_ids=chunk_ids,
)
raw_facts = await extract_facts_from_transcript_async(
llm, numbered, narrator_name=narrator_name
)
seen: set[tuple] = set()
inserted: list[dict] = []
for f in raw_facts:
key = dedupe_key(f, narrator_name=narrator_name)
if key in seen:
continue
seen.add(key)
scid = f.get("source_chunk_id")
if scid and scid not in chunk_ids:
scid = chunk_ids[0] if chunk_ids else None
row = await create_memory_fact(
db,
user_id=user_id,
fact_type=f.get("fact_type") or "event",
subject=normalize_subject(f.get("subject"), narrator_name),
predicate=f.get("predicate"),
object_json=normalize_object_json(f.get("object_json")),
confidence=float(f.get("confidence") or 0.75),
source_chunk_id=scid,
status="confirmed",
)
inserted.append(
{
"id": row.id,
"fact_type": row.fact_type,
"subject": row.subject,
"predicate": row.predicate,
"object_json": row.object_json,
}
)
if inserted:
await delete_timeline_events_by_memory_source(
db, user_id=user_id, memory_source_id=source_id
)
events = await build_timeline_events_from_facts_async(llm, inserted)
for ev in events:
await create_timeline_event(
db,
user_id=user_id,
event_year=ev.get("event_year"),
event_date=ev.get("event_date"),
title=ev["title"],
description=ev.get("description"),
source_fact_ids=ev.get("source_fact_ids") or None,
memory_source_id=source_id,
)