退货欺诈检测标签引擎 — 识别假退货/滥用退货政策的行为模式与Tag自动标记

"""
退货欺诈检测标签引擎
功能：多维度行为特征提取 / 欺诈评分 / Tag生成 / 自动审核建议
"""
import numpy as np
from dataclasses import dataclass, field
from datetime import datetime, timedelta
import warnings
warnings.filterwarnings('ignore')


@dataclass
class ReturnRequest:
    request_id: str
    customer_id: str
    sku_id: str
    order_date: datetime
    return_date: datetime
    claimed_reason: str
    return_weight_kg: float
    original_weight_kg: float
    order_value_usd: float
    account_age_days: int
    customer_return_count_30d: int
    customer_complaint_count_30d: int
    tags: dict = field(default_factory=dict)


def compute_fraud_score(req: ReturnRequest) -> dict:
    signals = []
    score = 0.0

    # 信号1：高频退货
    if req.customer_return_count_30d >= 5:
        score += 0.35
        signals.append(f"高频退货({req.customer_return_count_30d}次/30天)")
    elif req.customer_return_count_30d >= 3:
        score += 0.20
        signals.append(f"频繁退货({req.customer_return_count_30d}次/30天)")

    # 信号2：重量异常
    weight_ratio = req.return_weight_kg / max(0.01, req.original_weight_kg)
    if weight_ratio < 0.75:
        score += 0.30
        signals.append(f"重量异常({weight_ratio:.0%}，疑似空包)")
    elif weight_ratio < 0.90:
        score += 0.15
        signals.append(f"重量偏低({weight_ratio:.0%})")

    # 信号3：新账号+高价值退货
    if req.account_age_days < 30 and req.order_value_usd > 100:
        score += 0.25
        signals.append(f"新账号({req.account_age_days}天)+高价值(${req.order_value_usd:.0f})")

    # 信号4：频繁投诉
    if req.customer_complaint_count_30d >= 3:
        score += 0.15
        signals.append(f"投诉频繁({req.customer_complaint_count_30d}次/30天)")

    # 信号5：使用周期末退货（季节性）