from __future__ import annotations import json import re from collections import Counter from datetime import datetime, timedelta, timezone from pathlib import Path from dateutil import parser as date_parser from pydantic import ValidationError from app.core.config import get_settings from app.core.logging import logger from app.domain.schemas import ExtractResponse, JobCard, Salary, SkillScore, WorkerCard from app.services.llm_client import LLMClient from app.utils.ids import generate_id from app.utils.prompts import load_prompt class ExtractionService: def __init__(self) -> None: self.settings = get_settings() self.skills = json.loads((self.settings.sample_data_dir / "skills.json").read_text(encoding="utf-8")) self.categories = json.loads((self.settings.sample_data_dir / "categories.json").read_text(encoding="utf-8")) self.regions = json.loads((self.settings.sample_data_dir / "regions.json").read_text(encoding="utf-8")) self.sample_jobs = json.loads((self.settings.sample_data_dir / "jobs.json").read_text(encoding="utf-8")) self.sample_workers = json.loads((self.settings.sample_data_dir / "workers.json").read_text(encoding="utf-8")) self.default_region = self._build_default_region() self.default_category = self._build_default_category() self.default_salary_amount = self._build_default_salary_amount() self.default_job_tags = self._build_default_job_tags() self.default_worker_skills = self._build_default_worker_skills() self.default_experience_tags = self._build_default_experience_tags() self.category_skill_defaults = self._build_category_skill_defaults() self.city_region_defaults = self._build_city_region_defaults() self.tag_candidates = self._build_tag_candidates() self.llm_client = LLMClient(self.settings) self.shanghai_tz = timezone(timedelta(hours=8)) def extract_job(self, text: str) -> ExtractResponse: logger.info("extract_job request text=%s", text) llm_card = self._llm_extract_with_retry(text, self.settings.prompt_dir / "job_extract.md", JobCard) if llm_card: return ExtractResponse(success=True, data=llm_card) try: card = self._extract_job_rule(text) return ExtractResponse(success=True, data=card) except ValidationError as exc: logger.exception("Rule job extraction validation failed") return ExtractResponse(success=False, errors=[str(exc)], missing_fields=self._missing_fields(exc)) def extract_worker(self, text: str) -> ExtractResponse: logger.info("extract_worker request text=%s", text) llm_card = self._llm_extract_with_retry(text, self.settings.prompt_dir / "worker_extract.md", WorkerCard) if llm_card: return ExtractResponse(success=True, data=llm_card) try: card = self._extract_worker_rule(text) return ExtractResponse(success=True, data=card) except ValidationError as exc: logger.exception("Rule worker extraction validation failed") return ExtractResponse(success=False, errors=[str(exc)], missing_fields=self._missing_fields(exc)) def _llm_extract(self, text: str, prompt_path: Path): try: return self.llm_client.extract_json(load_prompt(prompt_path), text) except Exception: logger.exception("LLM extraction failed, fallback to rule-based extraction") return None def _llm_extract_with_retry(self, text: str, prompt_path: Path, schema_cls): base_prompt = load_prompt(prompt_path) llm_result = self._llm_extract(text, prompt_path) if not llm_result: return None try: return schema_cls(**llm_result.content) except ValidationError as exc: logger.warning("LLM extraction validation failed, trying schema-aware retry") last_error = exc last_output = llm_result.content for _ in range(self.settings.extraction_llm_max_retries): missing_fields = self._missing_fields(last_error) repair_prompt = self._build_repair_prompt(base_prompt, schema_cls, missing_fields) try: repair_result = self.llm_client.extract_json( repair_prompt, self._build_repair_input(text, last_output, missing_fields), ) except Exception: logger.exception("LLM schema-aware retry failed") return None if not repair_result: return None last_output = repair_result.content try: return schema_cls(**repair_result.content) except ValidationError as exc: last_error = exc logger.warning("LLM schema-aware retry still invalid missing_fields=%s", self._missing_fields(exc)) return None def _build_repair_prompt(self, base_prompt: str, schema_cls, missing_fields: list[str]) -> str: schema_json = json.dumps(schema_cls.model_json_schema(), ensure_ascii=False) return ( f"{base_prompt}\n\n" "你是结构化修复助手。请严格输出可被 JSON 解析的对象，不要输出解释文字。\n" "目标是根据给定 schema 修复字段缺失和类型错误，优先保证必填字段完整。\n" f"缺失或错误字段: {', '.join(missing_fields) if missing_fields else 'unknown'}\n" f"JSON Schema: {schema_json}\n" ) def _build_repair_input(self, original_text: str, last_output: dict, missing_fields: list[str]) -> str: return ( f"原始文本:\n{original_text}\n\n" f"上一次抽取结果:\n{json.dumps(last_output, ensure_ascii=False)}\n\n" f"请重点修复字段:\n{json.dumps(missing_fields, ensure_ascii=False)}" ) def _extract_job_rule(self, text: str) -> JobCard: skill_hits = [item for item in self.skills if item in text] category = next((item for item in self.categories if item in text), self.default_category) region = self._extract_region(text) salary = self._extract_salary(text) headcount = self._extract_number(text, [r"(\d+)\s*[个名人位]"], default=1) duration = self._extract_number(text, [r"(\d+(?:\.\d+)?)\s*小时"], default=4.0, cast=float) tags = [tag for tag in self.tag_candidates if tag in text][:3] title = next((f"{category}{skill_hits[0]}兼职" for _ in [0] if skill_hits), f"{category}兼职") card = JobCard( job_id=generate_id("job"), title=title, category=category, description=text, skills=skill_hits[:5] or self._guess_category_skills(category), city=region["city"], region=region["region"], location_detail=self._extract_location(text, region), start_time=self._extract_job_time(text), duration_hours=duration, headcount=int(headcount), salary=salary, work_mode="排班制" if "排班" in text else "兼职", tags=tags or self.default_job_tags, confidence=self._compute_confidence(skill_hits, region, salary.amount > 0), ) return card def _extract_worker_rule(self, text: str) -> WorkerCard: skill_hits = [item for item in self.skills if item in text][:6] region_hits = [item for item in self.regions if item["region"] in text or item["city"] in text] if not region_hits: city_hits = [item["city"] for item in self.regions if item["city"] in text] unique_city_hits = list(dict.fromkeys(city_hits)) region_hits = [ {"city": city, "region": self.city_region_defaults.get(city, self.default_region["region"])} for city in unique_city_hits ] city_names = list(dict.fromkeys([item["city"] for item in region_hits])) or [self.default_region["city"]] region_names = list(dict.fromkeys([item["region"] for item in region_hits])) or [self.default_region["region"]] availability = self._extract_availability(text) experience = [item for item in self.default_experience_tags if item in text] card = WorkerCard( worker_id=generate_id("worker"), name=self._extract_name(text), description=text, skills=[ SkillScore(name=item, score=round(0.72 + index * 0.04, 2)) for index, item in enumerate(skill_hits or self.default_worker_skills) ], cities=city_names, regions=region_names, availability=availability, experience_tags=experience or self.default_experience_tags[:2], reliability_score=0.76, profile_completion=0.68, confidence=self._compute_confidence(skill_hits, {"city": city_names[0], "region": region_names[0]}, True), ) return card def _extract_region(self, text: str) -> dict: for item in self.regions: if item["city"] in text and item["region"] in text: return item for item in self.regions: if item["region"] in text: return item city_match = next((item["city"] for item in self.regions if item["city"] in text), "") if city_match: return {"city": city_match, "region": self.city_region_defaults.get(city_match, self.default_region["region"])} return self.default_region def _extract_location(self, text: str, region: dict) -> str: markers = ["会展中心", "商场", "地铁站", "园区", "写字楼", "仓库", "门店"] for marker in markers: if marker in text: return f"{region['city']}{region['region']}{marker}" return f"{region['city']}{region['region']}待定点位" def _extract_salary(self, text: str) -> Salary: amount = self._extract_number(text, [r"(\d+(?:\.\d+)?)\s*(?:元|块)"], default=self.default_salary_amount, cast=float) salary_type = "hourly" if "小时" in text and "/小时" in text else "daily" return Salary(type=salary_type, amount=amount, currency="CNY") def _extract_number(self, text: str, patterns: list[str], default, cast=int): for pattern in patterns: match = re.search(pattern, text) if match: return cast(match.group(1)) return default def _extract_job_time(self, text: str) -> datetime: now = datetime.now(self.shanghai_tz) for candidate in self._time_candidates(text, now): parsed = self._parse_datetime(candidate, now) if parsed: return parsed return self._normalize_datetime(now + timedelta(days=1)) def _time_candidates(self, text: str, now: datetime) -> list[str]: candidates = [text] if any(token in text for token in ("今天", "今日")): candidates.append(text.replace("今日", now.strftime("%Y-%m-%d")).replace("今天", now.strftime("%Y-%m-%d"))) if "明天" in text: tomorrow = now + timedelta(days=1) candidates.append(text.replace("明天", tomorrow.strftime("%Y-%m-%d"))) if "后天" in text: day_after = now + timedelta(days=2) candidates.append(text.replace("后天", day_after.strftime("%Y-%m-%d"))) weekday_map = {"一": 0, "二": 1, "三": 2, "四": 3, "五": 4, "六": 5, "日": 6, "天": 6} week_match = re.search(r"(下周|本周|这周|周)([一二三四五六日天])", text) if week_match: week_token, weekday_token = week_match.groups() target_weekday = weekday_map[weekday_token] days_ahead = (target_weekday - now.weekday()) % 7 if week_token == "下周": days_ahead = days_ahead + 7 elif week_token == "周" and days_ahead == 0: days_ahead = 7 target_day = now + timedelta(days=days_ahead) candidates.append(text.replace(week_match.group(0), target_day.strftime("%Y-%m-%d"))) return candidates def _parse_datetime(self, text: str, now: datetime) -> datetime | None: normalized = self._replace_time_words(text) cleaned = re.sub(r"[，、。；,;]", " ", normalized) cleaned = cleaned.replace("号", "日") cleaned = re.sub(r"(\d{1,2})月(\d{1,2})日", rf"{now.year}-\1-\2", cleaned) cleaned = re.sub(r"(\d{1,2})点半", r"\1:30", cleaned) cleaned = re.sub(r"(\d{1,2})点", r"\1:00", cleaned) cleaned = re.sub(r"(\d{1,2})时", r"\1:00", cleaned) has_date = bool(re.search(r"\d{4}-\d{1,2}-\d{1,2}", cleaned)) if not has_date: return None try: parsed = date_parser.parse(cleaned, fuzzy=True) except Exception: return None return self._normalize_datetime(parsed) def _replace_time_words(self, text: str) -> str: replaced = text replaced = re.sub(r"(今晚|晚上)", " 19:00 ", replaced) replaced = re.sub(r"(下午)", " 14:00 ", replaced) replaced = re.sub(r"(中午)", " 12:00 ", replaced) replaced = re.sub(r"(早上|上午)", " 09:00 ", replaced) replaced = re.sub(r"(凌晨)", " 01:00 ", replaced) return replaced def _normalize_datetime(self, value: datetime) -> datetime: if value.tzinfo is None: value = value.replace(tzinfo=self.shanghai_tz) else: value = value.astimezone(self.shanghai_tz) return value.replace(second=0, microsecond=0) def _extract_availability(self, text: str) -> list[str]: tags = [] if "周末" in text: tags.append("weekend") if "上午" in text: tags.append("weekday_am") if "下午" in text: tags.append("weekday_pm") if "随时" in text or "都能" in text or "全天" in text: tags.append("anytime") return tags or ["anytime"] def _extract_name(self, text: str) -> str: if match := re.search(r"我叫([\u4e00-\u9fa5]{2,4})", text): return match.group(1) if match := re.search(r"我是([\u4e00-\u9fa5]{2,4})", text): return match.group(1) return "匿名候选人" def _guess_category_skills(self, category: str) -> list[str]: skills = self.category_skill_defaults.get(category) if skills: return skills return self.default_worker_skills[:3] def _compute_confidence(self, skill_hits: list[str], region: dict, has_salary: bool) -> float: score = 0.55 if skill_hits: score += 0.15 if region.get("city"): score += 0.15 if has_salary: score += 0.1 return min(round(score, 2), 0.95) def _missing_fields(self, exc: ValidationError) -> list[str]: return [".".join(str(part) for part in item["loc"]) for item in exc.errors()] def _build_default_region(self) -> dict: if self.sample_jobs: pair_counter = Counter( (item.get("city"), item.get("region")) for item in self.sample_jobs if item.get("city") and item.get("region") ) if pair_counter: city, region = pair_counter.most_common(1)[0][0] return {"city": city, "region": region} if self.regions: return {"city": self.regions[0]["city"], "region": self.regions[0]["region"]} return {"city": "深圳", "region": "南山"} def _build_default_category(self) -> str: counter = Counter(item.get("category") for item in self.sample_jobs if item.get("category")) if counter: return counter.most_common(1)[0][0] return self.categories[0] if self.categories else "活动执行" def _build_default_salary_amount(self) -> float: amounts = sorted( float(item["salary"]["amount"]) for item in self.sample_jobs if isinstance(item.get("salary"), dict) and isinstance(item["salary"].get("amount"), (int, float)) ) if not amounts: return 150.0 mid = len(amounts) // 2 if len(amounts) % 2 == 1: return amounts[mid] return round((amounts[mid - 1] + amounts[mid]) / 2, 2) def _build_default_job_tags(self) -> list[str]: counter = Counter( tag for item in self.sample_jobs for tag in item.get("tags", []) if isinstance(tag, str) and tag.strip() ) top_tags = [tag for tag, _ in counter.most_common(3)] return top_tags or ["有经验优先"] def _build_default_worker_skills(self) -> list[str]: counter = Counter( skill.get("name") for item in self.sample_workers for skill in item.get("skills", []) if isinstance(skill, dict) and isinstance(skill.get("name"), str) and skill.get("name") ) top_skills = [name for name, _ in counter.most_common(4)] return top_skills or ["活动执行", "引导", "登记"] def _build_default_experience_tags(self) -> list[str]: counter = Counter( tag for item in self.sample_workers for tag in item.get("experience_tags", []) if isinstance(tag, str) and tag.strip() ) top_tags = [tag for tag, _ in counter.most_common(5)] return top_tags or ["活动执行"] def _build_category_skill_defaults(self) -> dict[str, list[str]]: category_skills: dict[str, Counter] = {} for item in self.sample_jobs: category = item.get("category") if not isinstance(category, str) or not category: continue counter = category_skills.setdefault(category, Counter()) for skill in item.get("skills", []): if isinstance(skill, str) and skill: counter[skill] += 1 return {category: [name for name, _ in counter.most_common(4)] for category, counter in category_skills.items()} def _build_city_region_defaults(self) -> dict[str, str]: counter: dict[str, Counter] = {} for item in self.regions: city = item.get("city") region = item.get("region") if not city or not region: continue counter.setdefault(city, Counter())[region] += 1 for item in self.sample_jobs: city = item.get("city") region = item.get("region") if city and region: counter.setdefault(city, Counter())[region] += 3 defaults: dict[str, str] = {} for city, regions in counter.items(): defaults[city] = regions.most_common(1)[0][0] return defaults def _build_tag_candidates(self) -> list[str]: sample_tags = list( dict.fromkeys( tag for item in self.sample_jobs for tag in item.get("tags", []) if isinstance(tag, str) and tag.strip() ) ) baseline_tags = ["女生优先", "男生优先", "有经验优先", "沟通好", "可连做优先"] merged = list(dict.fromkeys([*sample_tags, *baseline_tags])) return merged[:30]