import json import os import logging import httpx from app import database logger = logging.getLogger(__name__) _985_UNIVERSITIES = { "清华大学","北京大学","复旦大学","上海交通大学","浙江大学","南京大学","中国科学技术大学", "哈尔滨工业大学","西安交通大学","武汉大学","华中科技大学","中山大学","四川大学","南开大学", "天津大学","山东大学","东南大学","吉林大学","厦门大学","同济大学","北京师范大学", "国防科技大学","中国人民大学","北京航空航天大学","北京理工大学","中国农业大学", "中央民族大学","华东师范大学","兰州大学","西北工业大学","电子科技大学", "重庆大学","湖南大学","中南大学","大连理工大学","东北大学","华南理工大学" } _211_UNIVERSITIES = _985_UNIVERSITIES | { "北京科技大学","北京交通大学","北京邮电大学","北京林业大学","北京外国语大学", "中国传媒大学","中央财经大学","对外经济贸易大学","北京体育大学","中国政法大学", "华北电力大学","河北工业大学","太原理工大学","内蒙古大学","辽宁大学","大连海事大学", "延边大学","东北师范大学","哈尔滨工程大学","东北农业大学","华东理工大学","东华大学", "上海外国语大学","上海财经大学","上海大学","苏州大学","南京航空航天大学","南京理工大学", "中国矿业大学","河海大学","江南大学","南京农业大学","中国药科大学","南京师范大学", "安徽大学","合肥工业大学","福州大学","南昌大学","中国石油大学","郑州大学", "中国地质大学","武汉理工大学","华中农业大学","华中师范大学","中南财经政法大学", "湖南师范大学","暨南大学","华南师范大学","广西大学","海南大学","西南交通大学", "西南大学","西南财经大学","贵州大学","云南大学","西藏大学","西北大学", "西安电子科技大学","长安大学","陕西师范大学","青海大学","宁夏大学","石河子大学", "新疆大学","北京工业大学","天津医科大学","南京信息工程大学","南方科技大学", "上海科技大学","中国社科院大学","中国科学院大学" } def _get_school_tier(school: str) -> int: if not school: return 0 if school in _985_UNIVERSITIES: return 2 if school in _211_UNIVERSITIES: return 1 return 0 def _build_resume_text(resume_data: dict) -> str: skills = " ".join(s.lower().strip() for s in resume_data.get("skills", [])) projects = " ".join( p if isinstance(p, str) else (p.get("name", "") + " " + p.get("description", "")) for p in resume_data.get("projects", []) ) parts = [ resume_data.get("summary") or "", projects, resume_data.get("experience") or "", resume_data.get("education") or "", resume_data.get("major") or "", skills, ] return " ".join(parts) def _calc_match_rate(resume_data: dict, job: dict, tag_score: float, edu_score: float, vec_score: float) -> float: final_score = vec_score * 0.35 + tag_score * 0.45 + edu_score * 0.20 return round(min(final_score, 1.0), 4) def _calc_tag_score(resume_data: dict, job: dict) -> float: skills = [s.lower().strip() for s in resume_data.get("skills", [])] tags = [t.lower().strip() for t in job.get("tags", [])] if not tags: return 0.0 matched = sum(1 for t in tags if t in skills) return matched / len(tags) def _calc_edu_score(resume_data: dict, job: dict) -> float: school = (resume_data.get("school") or "") grade = (resume_data.get("grade") or "").lower() jd_requirements = (job.get("requirements") or "").lower() score = 0.4 tier = _get_school_tier(school) if tier >= 2: score = 1.0 elif tier >= 1: score = 0.8 elif school: score = 0.5 if ("硕士" in grade or "博士" in grade or "研" in grade) and "硕士" in jd_requirements: score = min(score + 0.15, 1.0) return score def recommend_jobs(resume_data: dict) -> list[dict]: if not resume_data or not resume_data.get("skills"): return [] conn = database.get_conn() rows = conn.execute("SELECT * FROM jobs ORDER BY job_id").fetchall() conn.close() jobs = [] for r in rows: embedding_raw = r["embedding"] if "embedding" in r.keys() else "" jobs.append({ "job_id": r["job_id"], "name": r["name"], "company": r["company"], "city": r["city"], "salary": r["salary"], "tags": json.loads(r["tags"] or "[]"), "description": r["description"], "requirements": r["requirements"], "expectation_signals": json.loads(r["expectation_signals"] or "[]"), "_embedding": json.loads(embedding_raw) if embedding_raw else None, }) resume_text = _build_resume_text(resume_data) resume_vec = None try: from app.services.embedding_service import compute_embedding, cosine_similarity resume_vec = compute_embedding(resume_text) except Exception as e: logger.warning("Resume embedding failed: %s", e) for job in jobs: tag_score = _calc_tag_score(resume_data, job) edu_score = _calc_edu_score(resume_data, job) vec_raw = 0.0 if resume_vec and job["_embedding"]: vec_raw = cosine_similarity(resume_vec, job["_embedding"]) job["_tag_score"] = tag_score job["_edu_score"] = edu_score job["_vec_raw"] = vec_raw for job in jobs: job["_vec_score"] = max(0, min((job["_vec_raw"] - 0.30) / 0.40, 1.0)) scored = [] for job in jobs: match_rate = _calc_match_rate( resume_data, job, job["_tag_score"], job["_edu_score"], job["_vec_score"] ) scored.append({ "job_id": job["job_id"], "match_rate": match_rate, "job": job, }) scored.sort(key=lambda x: x["match_rate"], reverse=True) top_n = scored api_key = os.getenv("DEEPSEEK_API_KEY") if api_key and top_n: try: job_summaries = [] for item in top_n: j = item["job"] job_summaries.append( f' {j.get("job_id")}: {j.get("name")} @ {j.get("company")} | ' f'tags={j.get("tags")} | match_rate={item["match_rate"]}' ) system_prompt = ( "你是一个岗位推荐排序专家。给定候选人的技能列表和全部候选岗位," "请根据岗位与候选人的匹配程度重新排序,输出最匹配的岗位ID列表(从最匹配到最不匹配)。" "输出格式为JSON:{\"ranked_job_ids\": [1, 2, 3, ...]}" ) user_input = { "skills": resume_data.get("skills", []), "candidates": "\n".join(job_summaries), } with httpx.Client(timeout=30.0) as client: resp = client.post( "https://api.deepseek.com/chat/completions", headers={ "Authorization": f"Bearer {api_key}", "Content-Type": "application/json", }, json={ "model": "deepseek-chat", "response_format": {"type": "json_object"}, "messages": [ {"role": "system", "content": system_prompt}, {"role": "user", "content": json.dumps(user_input, ensure_ascii=False)}, ], }, ) resp.raise_for_status() body = resp.json() content = body["choices"][0]["message"]["content"] llm_result = json.loads(content) ranked_ids = llm_result.get("ranked_job_ids", []) id_to_item = {item["job"]["job_id"]: item for item in top_n} re_ranked = [] for raw_id in ranked_ids: if raw_id in id_to_item: re_ranked.append(id_to_item[raw_id]) for item in top_n: if item["job"]["job_id"] not in ranked_ids: re_ranked.append(item) top_n = re_ranked except Exception as e: logger.warning("LLM ranking failed: %s", e) return [ { "job_id": item["job_id"], "match_rate": item["match_rate"], "match_detail": { "semantic": round(item["job"]["_vec_score"], 4), "tags": round(item["job"]["_tag_score"], 4), "education": round(item["job"]["_edu_score"], 4), }, "name": item["job"].get("name", ""), "company": item["job"].get("company", ""), "city": item["job"].get("city", ""), "salary": item["job"].get("salary", ""), "tags": item["job"].get("tags", []), "description": item["job"].get("description", ""), "requirements": item["job"].get("requirements", ""), "expectation_signals": item["job"].get("expectation_signals", []), } for item in top_n ]