import re REGION_TOKENS = [ "서울", "경기", "인천", "수원", "용인", "성남", "화성", "동탄", "오산", "평택", "안양", "안산", "과천", "광명", "시흥", "하남", "남양주", "김포", "파주", "고양", "의정부", "대전", "대구", "부산", "광주", "울산", "세종", "제주" ] PROPERTY_TOKENS = [ "아파트", "오피스텔", "빌라", "주택", "상가", "토지", "원룸", "투룸", "다가구" ] TRANSACTION_TOKENS = [ "매매", "전세", "월세", "분양", "임대", "급매", "투자", "실거주", "매물" ] INFO_TOKENS = [ "시세", "학군", "입지", "교통", "생활권", "호재", "비교", "분석", "체크", "정리", "추천" ] APT_BRAND_TOKENS = [ "자이", "래미안", "푸르지오", "더샵", "힐스테이트", "롯데캐슬", "아이파크", "e편한", "리슈빌", "센트럴", "파크", "캐슬", "하이츠", "SK뷰", "두산위브", "포레나", "베르디움", "디에트르", "금호어울림", "하늘채", "데시앙", "엘리프", "제일풍경채", "서희", "센트레빌", "한신더휴", "스위첸", "듀크", "우미", "비발디", "유보라", "뜰", "플래티넘", "빌리브", "애시앙", "펜테리움", "헤링턴", "더리브", "파크드림" ] def clean_spaces(text: str) -> str: return re.sub(r"\s+", " ", text or "").strip() def extract_first_token(text: str, tokens: list[str]) -> str: for token in tokens: if token in text: return token return "" def analyze_title_pattern(title: str) -> dict: t = clean_spaces(title) region = extract_first_token(t, REGION_TOKENS) prop = extract_first_token(t, PROPERTY_TOKENS) txn = extract_first_token(t, TRANSACTION_TOKENS) info = extract_first_token(t, INFO_TOKENS) brand = extract_first_token(t, APT_BRAND_TOKENS) return { "title": t, "has_region": bool(region), "has_property": bool(prop), "has_transaction": bool(txn), "has_info": bool(info), "has_brand": bool(brand), "region": region, "property": prop, "transaction": txn, "info": info, "brand": brand, "length": len(t), } def summarize_patterns(posts: list[dict], max_items: int = 10) -> dict: titles = [clean_spaces(p.get("title", "")) for p in (posts or [])[:max_items] if clean_spaces(p.get("title", ""))] patterns = [analyze_title_pattern(t) for t in titles] if not patterns: return { "count": 0, "avg_length": 0, "region_count": 0, "property_count": 0, "transaction_count": 0, "info_count": 0, "brand_count": 0, "top_regions": [], "top_transactions": [], "top_properties": [], "sample_titles": [], } def top_values(key: str, limit: int = 3): counts = {} for p in patterns: val = p.get(key, "") if val: counts[val] = counts.get(val, 0) + 1 return sorted(counts.items(), key=lambda x: (-x[1], x[0]))[:limit] return { "count": len(patterns), "avg_length": round(sum(p["length"] for p in patterns) / len(patterns), 2), "region_count": sum(1 for p in patterns if p["has_region"]), "property_count": sum(1 for p in patterns if p["has_property"]), "transaction_count": sum(1 for p in patterns if p["has_transaction"]), "info_count": sum(1 for p in patterns if p["has_info"]), "brand_count": sum(1 for p in patterns if p["has_brand"]), "top_regions": top_values("region"), "top_transactions": top_values("transaction"), "top_properties": top_values("property"), "sample_titles": titles[:5], } def pattern_comment(base_summary: dict, comp_summary: dict, comp_blog_id: str) -> list[str]: comments = [] if comp_summary["region_count"] > base_summary["region_count"]: comments.append(f"{comp_blog_id}는 제목에 지역명을 더 자주 넣고 있습니다.") if comp_summary["transaction_count"] > base_summary["transaction_count"]: comments.append(f"{comp_blog_id}는 매매·전세·월세 같은 거래형 키워드를 더 자주 사용합니다.") if comp_summary["property_count"] > base_summary["property_count"]: comments.append(f"{comp_blog_id}는 아파트·오피스텔 등 매물유형 표현이 더 분명합니다.") if comp_summary["info_count"] > base_summary["info_count"]: comments.append(f"{comp_blog_id}는 시세·입지·분석형 제목 비중이 더 높습니다.") if comp_summary["brand_count"] > base_summary["brand_count"]: comments.append(f"{comp_blog_id}는 단지명 또는 브랜드명 중심 제목을 더 자주 사용합니다.") if comp_summary["avg_length"] > base_summary["avg_length"] + 4: comments.append(f"{comp_blog_id}는 제목 길이가 조금 더 길고 구체적인 편입니다.") return comments[:5] def generate_pattern_based_titles(base_posts: list[dict], comp_posts: list[dict], max_items: int = 3) -> list[str]: base_titles = [clean_spaces(p.get("title", "")) for p in (base_posts or []) if clean_spaces(p.get("title", ""))] comp_titles = [clean_spaces(p.get("title", "")) for p in (comp_posts or []) if clean_spaces(p.get("title", ""))] base_join = " ".join(base_titles) comp_join = " ".join(comp_titles) region = extract_first_token(base_join, REGION_TOKENS) or extract_first_token(comp_join, REGION_TOKENS) or "지역" prop = extract_first_token(base_join, PROPERTY_TOKENS) or extract_first_token(comp_join, PROPERTY_TOKENS) or "아파트" txn = extract_first_token(base_join, TRANSACTION_TOKENS) or extract_first_token(comp_join, TRANSACTION_TOKENS) or "매매" info = extract_first_token(comp_join, INFO_TOKENS) or "시세" brand = extract_first_token(base_join, APT_BRAND_TOKENS) or extract_first_token(comp_join, APT_BRAND_TOKENS) suggestions = [] if brand: suggestions.append(f"{region} {brand} {prop} {txn} {info} 총정리") suggestions.append(f"{region} {brand} {prop} {txn} 체크포인트") suggestions.append(f"{region} {brand} {prop} {txn} 추천 이유") else: suggestions.append(f"{region} {prop} {txn} {info} 총정리") suggestions.append(f"{region} {prop} {txn} 입지와 체크포인트") suggestions.append(f"{region} {prop} {txn} 추천 이유와 비교") out = [] seen = set() for s in suggestions: s = clean_spaces(s) if s and s not in seen: seen.add(s) out.append(s) return out[:max_items] def build_title_pattern_insights(results: list[dict], base_blog_id: str = "") -> list[dict]: successful = [x for x in results if x.get("success")] if len(successful) < 2: return [] base = None if base_blog_id: for item in successful: if item.get("blog", {}).get("blog_id", "") == base_blog_id: base = item break if base is None: base = successful[0] base_posts = base.get("posts", []) or [] base_summary = summarize_patterns(base_posts, 10) base_blog_id_final = base.get("blog", {}).get("blog_id", "") competitors = [ x for x in successful if x.get("blog", {}).get("blog_id", "") != base_blog_id_final ] insights = [] for comp in competitors: comp_blog_id = comp.get("blog", {}).get("blog_id", "") comp_posts = comp.get("posts", []) or [] comp_summary = summarize_patterns(comp_posts, 10) insights.append({ "base_blog_id": base_blog_id_final, "competitor_blog_id": comp_blog_id, "base_pattern_summary": base_summary, "competitor_pattern_summary": comp_summary, "pattern_comments": pattern_comment(base_summary, comp_summary, comp_blog_id), "recommended_titles": generate_pattern_based_titles(base_posts, comp_posts, 3), }) return insights