labor-arbitration-system/backend/app/services/portrait_generator.py

from __future__ import annotations

import math
import re
from typing import Any


def _clamp(v: float, lo: float = 0, hi: float = 100) -> int:
    return int(max(lo, min(hi, v)))


def _has(v: Any) -> bool:
    if v is None:
        return False
    if isinstance(v, str):
        return bool(v.strip())
    if isinstance(v, (list, dict)):
        return len(v) > 0
    return True


def _keyword_tags(elements: dict[str, Any]) -> list[dict[str, Any]]:
    """
    多级标签体系（MVP）：一级维度 + 二级标签 + 具体值
    """
    cause = elements.get("case_cause")
    tags: list[dict[str, Any]] = []
    if cause:
        tags.append(
            {
                "level1": "法律维度",
                "level2": "争议焦点类型",
                "value": str(cause),
            }
        )
    if elements.get("overtime_desc"):
        tags.append({"level1": "事实维度", "level2": "关键事实", "value": "加班事实"})
    if elements.get("termination_reason"):
        tags.append({"level1": "事实维度", "level2": "解除情形", "value": "已识别解除原因"})
    return tags


def _extract_keywords(text: str, top_k: int = 15) -> list[dict[str, Any]]:
    """
    用 jieba 分词提取高频关键词，返回连贯的中文词语。
    """
    if not text:
        return []
    try:
        import jieba
        # 添加法律领域自定义词，避免复合词被切分
        for w in ["劳动合同", "经济补偿", "赔偿金", "拖欠工资", "违法解除",
                   "加班费", "工伤认定", "社会保险", "住房公积金",
                   "解除劳动合同", "双倍工资", "违法辞退", "劳动仲裁",
                   "科技有限公司", "有限公司", "劳动争议", "仲裁委员会",
                   "年终奖", "二倍工资"]:
            jieba.add_word(w)
        words = jieba.cut(text[:12000])
        toks = [w.strip() for w in words if len(w.strip()) >= 2]
    except ImportError:
        toks = re.findall(r"[一-鿿]{2,6}|[A-Za-z]{3,}", text[:12000])

    stop = {
        "申请人", "被申请人", "仲裁", "请求", "事实", "理由", "事项",
        "劳动", "争议", "委员会", "申请", "本案", "仲裁庭",
        "经审理", "查明", "如下", "予以", "认定",
        "支持", "驳回", "维持", "的", "了", "在", "和",
        "是", "不", "与", "及", "年", "月", "日", "元",
        "支付", "公司", "人民", "法院", "原告", "被告",
    }
    freq: dict[str, int] = {}
    for t in toks:
        if t in stop or len(t) <= 1:
            continue
        freq[t] = freq.get(t, 0) + 1
    items = sorted(freq.items(), key=lambda x: x[1], reverse=True)[:top_k]
    return [{"name": k, "value": v} for k, v in items]

def generate_portrait(elements: dict[str, Any], raw_text: str = "", evidence_count: int = 0) -> dict[str, Any]:
    laws = elements.get("law_refs") or []
    claims = elements.get("claims") or {}
    claim_items = claims.get("items") or []

    # 法律维度：争议焦点明确度 / 法条引用准确性(用“有无+数量”粗略替代) / 证据法律效力(用 evidence_count 近似)
    dispute_focus = 60 + (20 if _has(elements.get("case_cause")) else 0) + (10 if len(claim_items) >= 2 else 0)
    law_ref_score = 40 + min(40, 10 * len(laws))
    evidence_legal_power = 30 + min(50, evidence_count * 10)
    legal_score = _clamp(dispute_focus * 0.4 + law_ref_score * 0.35 + evidence_legal_power * 0.25)

    # 事实维度：时间线完整性 / 事实清晰度 / 证据完备度
    timeline_complete = 30 + (35 if _has(elements.get("entry_date")) else 0) + (35 if _has(elements.get("leave_date")) else 0)
    fact_clear = 40 + (25 if _has(elements.get("termination_reason")) else 0) + (20 if _has(elements.get("overtime_desc")) else 0)
    evidence_complete = 20 + min(60, evidence_count * 12)
    fact_score = _clamp(timeline_complete * 0.35 + fact_clear * 0.4 + evidence_complete * 0.25)

    # 风险维度：诉求支持可能性 / 矛盾激化程度
    amount_total = claims.get("amount_total")
    month_salary = elements.get("month_salary") or 0
    support_prob = 60
    if amount_total and month_salary:
        ratio = amount_total / max(month_salary, 1)
        if ratio > 12:
            support_prob -= 20
        elif ratio > 6:
            support_prob -= 10
        else:
            support_prob += 5
    if elements.get("case_cause") == "违法解除劳动合同" and not _has(elements.get("termination_reason")):
        support_prob -= 10
    support_prob = max(5, min(95, support_prob))

    escalation = 40 + min(40, len(claim_items) * 8) + (10 if (amount_total or 0) > 50000 else 0)
    risk_score = _clamp((100 - support_prob) * 0.6 + escalation * 0.4)

    # KPI：风险等级（给前端标签）
    risk_level = "低"
    if risk_score >= 70:
        risk_level = "高"
    elif risk_score >= 45:
        risk_level = "中"

    portrait = {
        "scores": {
            "legal": legal_score,
            "fact": fact_score,
            "risk": risk_score,
        },
        "legal_dimension": {
            "score": legal_score,
            "subscores": {
                "争议焦点明确度": _clamp(dispute_focus),
                "法律条款引用充分度": _clamp(law_ref_score),
                "证据法律效力(近似)": _clamp(evidence_legal_power),
            },
        },
        "fact_dimension": {
            "score": fact_score,
            "subscores": {
                "时间线完整性": _clamp(timeline_complete),
                "事实描述清晰度": _clamp(fact_clear),
                "证据完备度(近似)": _clamp(evidence_complete),
            },
        },
        "risk_dimension": {
            "score": risk_score,
            "subscores": {
                "诉求支持可能性(规则近似)": _clamp(support_prob),
                "矛盾激化程度(规则近似)": _clamp(escalation),
            },
        },
        "tags": _keyword_tags(elements),
        "keywords": _extract_keywords(raw_text),
        "risk_level": risk_level,
    }
    return portrait