MZC/KnowMathBot/backend/app/routers/recommendation.py

"""
个性化推荐路由：按知识点+权重组卷，或按错题优先组卷。
"""
from __future__ import annotations
import random
from itertools import groupby
from uuid import UUID
from fastapi import APIRouter, Depends
from sqlmodel import Session, select
from app.db import get_session
from app.deps import get_current_user
from app.models import Exercise, ExerciseAttempt, User, WrongBookItem
from app.schemas import RecommendItem, RecommendRequest, RecommendResponse
from app.utils.exercise_knowledge import (
    exercise_in_grade_category_scope,
    exercise_matches_any_selected,
)
from app.utils.knowledge_codes import split_knowledge_codes

router = APIRouter(prefix="/recommendation", tags=["recommendation"])

def _ex_weight(ex: Exercise) -> int:
    return int(getattr(ex, "weight", 0) or 0)

def _sort_by_weight_shuffle_ties(pool: list[Exercise]) -> list[Exercise]:
    pool_sorted = sorted(pool, key=_ex_weight, reverse=True)
    out: list[Exercise] = []
    for _, grp in groupby(pool_sorted, key=_ex_weight):
        chunk = list(grp)
        random.shuffle(chunk)
        out.extend(chunk)
    return out

def _recommend_topic_then_weight(
    all_exs: list[Exercise],
    selected_codes: list[str],
    limit: int,
    knowledge_grade: str | None,
    knowledge_category: str | None,
) -> list[RecommendItem]:
    """
    开始组卷：先限定在所选知识点（含同族编码），再按错题 weight 从高到低优先，同权重随机；
    不足时仅在「同年级+同类」内随机补足，避免全库混入其他大类。
    """
    # 第一步：只从“勾选知识点命中”的题里选。
    pool = [ex for ex in all_exs if exercise_matches_any_selected(ex, selected_codes)]
    ordered = _sort_by_weight_shuffle_ties(pool)
    picked: list[Exercise] = []
    picked_ids: set[UUID] = set()
    for ex in ordered:
        if len(picked) >= limit:
            break
        if ex.id not in picked_ids:
            picked.append(ex)
            picked_ids.add(ex.id)

    supplement_ids: set[UUID] = set()
    if len(picked) < limit:
        # 第二步：题量不足时，仅在同年级/同类别范围补足，防止跨类混题。
        g = (knowledge_grade or "").strip() or None
        c = (knowledge_category or "").strip() or None
        if g or c:
            rest = [
                ex
                for ex in all_exs
                if ex.id not in picked_ids and exercise_in_grade_category_scope(ex, g, c)
            ]
            random.shuffle(rest)
            for ex in rest:
                if len(picked) >= limit:
                    break
                picked.append(ex)
                picked_ids.add(ex.id)
                supplement_ids.add(ex.id)
        # 未传年级/类别时不再从全库随机补足（旧行为会混入几何等无关题）

    items: list[RecommendItem] = []
    for ex in picked:
        wv = _ex_weight(ex)
        if ex.id in supplement_ids:
            reason = "题量补足：同年级/同类随机（未混入其他类别）"
        else:
            reason = f"所选知识点；错题权重优先（weight={wv}）"
        items.append(RecommendItem(item_type="exercise", item_ref=str(ex.id), reason=reason))
    return items

def _recommend_personalized_wrong_then_random(
    session: Session,
    current_user: User,
    all_exs: list[Exercise],
    attempts: list[ExerciseAttempt],
    limit: int,
) -> list[RecommendItem]:
    """
    个性推荐组卷：不参考勾选知识点；优先 weight 高的错题，不足则随机补满。
    """
    # 错题来源 = 错题本 + 作答记录中判错题，综合反映“易错程度”。
    wrong_items = session.exec(select(WrongBookItem).where(WrongBookItem.user_id == current_user.id)).all()
    wrong_counts: dict[UUID, int] = {}
    for wb in wrong_items:
        wrong_counts[wb.exercise_id] = wrong_counts.get(wb.exercise_id, 0) + 1

    wrong_ex_ids: set[UUID] = set(wrong_counts.keys())
    wrong_ex_ids.update(a.exercise_id for a in attempts if a.is_correct is False)

    ex_by_id = {ex.id: ex for ex in all_exs}
    wrong_pool = [ex_by_id[eid] for eid in wrong_ex_ids if eid in ex_by_id]

    def sort_key(ex: Exercise) -> tuple[int, int]:
        return (-_ex_weight(ex), -wrong_counts.get(ex.id, 0))

    wrong_pool.sort(key=sort_key)
    # 同 (weight, wrong_count) 档内打乱，避免顺序固定
    out_wrong: list[Exercise] = []
    for _, grp in groupby(wrong_pool, key=sort_key):
        chunk = list(grp)
        random.shuffle(chunk)
        out_wrong.extend(chunk)

    picked: list[Exercise] = []
    picked_ids: set[UUID] = set()
    for ex in out_wrong:
        if len(picked) >= limit:
            break
        if ex.id not in picked_ids:
            picked.append(ex)
            picked_ids.add(ex.id)

    supplement_ids: set[UUID] = set()
    if len(picked) < limit:
        # 错题数量不够时，随机补足整卷题量。
        rest = [ex for ex in all_exs if ex.id not in picked_ids]
        random.shuffle(rest)
        for ex in rest:
            if len(picked) >= limit:
                break
            picked.append(ex)
            picked_ids.add(ex.id)
            supplement_ids.add(ex.id)

    items: list[RecommendItem] = []
    for ex in picked:
        wv = _ex_weight(ex)
        wc = wrong_counts.get(ex.id, 0)
        if ex.id in supplement_ids:
            reason = "个性推荐：题量随机补足"
        elif wc:
            reason = f"个性推荐：错题优先（weight={wv}，错题本{wc}条）"
        else:
            reason = f"个性推荐：错题优先（weight={wv}，曾答错）"
        items.append(RecommendItem(item_type="exercise", item_ref=str(ex.id), reason=reason))
    return items

@router.post("/recommend", response_model=RecommendResponse)
def recommend(payload: RecommendRequest, session: Session = Depends(get_session), current_user: User = Depends(get_current_user)):
    """组卷推荐：mode=topic_weight 按知识点+权重；personalized 按错题优先。"""
    limit = max(1, int(payload.limit or 10))
    raw_mode = (payload.mode or "").strip().lower()
    if raw_mode in ("topic_weight", "topic", "manual"):
        mode = "topic_weight"
    elif raw_mode in ("personalized", "personal"):
        mode = "personalized"
    else:
        # 兼容旧客户端：有知识点则按组卷，否则个性推荐
        mode = "topic_weight" if (payload.knowledge_code and payload.knowledge_code.strip()) else "personalized"

    all_exs = session.exec(select(Exercise)).all()
    attempts = session.exec(select(ExerciseAttempt).where(ExerciseAttempt.user_id == current_user.id)).all()

    if mode == "topic_weight":
        codes = split_knowledge_codes(payload.knowledge_code)
        if not codes:
            return RecommendResponse(items=[])
        items = _recommend_topic_then_weight(
            all_exs,
            codes,
            limit,
            payload.knowledge_grade,
            payload.knowledge_category,
        )
        return RecommendResponse(items=items)

    items = _recommend_personalized_wrong_then_random(session, current_user, all_exs, attempts, limit)
    return RecommendResponse(items=items)