Compare commits

..

No commits in common. "main" and "db-migration" have entirely different histories.

16 changed files with 113 additions and 268 deletions

2
.gitignore vendored
View File

@ -46,5 +46,3 @@ alembic/versions/*.pyc
test_results/
app/test*
docker-compose.yml

View File

@ -16,7 +16,7 @@ class FirecrawlClient:
HTTPMethod.POST,
url=f"{FIRECRAWL_BASE}/scrape",
headers=self._headers(),
json_body={"url": url, "formats": ["json", "links"], "jsonOptions": json_options, "waitFor": wait_for, "maxAge": 0},
json_body={"url": url, "formats": ["json", "links"], "jsonOptions": json_options, "waitFor": wait_for},
label="firecrawl-scrape",
)
if not resp or not resp.is_success:
@ -74,9 +74,9 @@ class FirecrawlClient:
headers=self._headers(),
json_body={
"url": url,
"formats": ["json", "links", "rawHtml"],
"formats": ["json", "links", "html"],
"jsonOptions": {
"prompt": "Extract: 클리닉 이름 - clinicName (Korean), clinic name (English), address, phone with dash format, business hours, slogan, services offered, doctors with name/title/specialty, brand identity (primary/accent/background/text colors in hex, heading/body fonts, logo URL from the actual header/main <img> src, og:image from <meta property='og:image'> content, favicon URL)",
"prompt": "Extract: clinic name (Korean), clinic name (English), address, phone with dash format, business hours, slogan, services offered, doctors with name/title/specialty, brand identity (primary/accent/background/text colors in hex, heading/body fonts, logo URL from the actual header/main <img> src, og:image from <meta property='og:image'> content, favicon URL)",
"schema": {
"type": "object",
"properties": {
@ -127,11 +127,8 @@ class FirecrawlClient:
},
},
"waitFor": 5000,
"maxAge": 0,
"proxy": "auto", # 기본 엔진이 차단되는 사이트(예: viewclinic.com)는 자동으로 stealth 프록시로 재시도.
"timeout": 120000, # proxy:auto면 60s로도 충분했지만(실측), 혹시 모르니 여유있게 120s.
},
timeout=150, # 위 Firecrawl 잡 타임아웃보다 길어야 우리 쪽 HTTP 클라이언트가 먼저 끊지 않음.
timeout=60,
label="firecrawl-clinic-info",
)
if not resp or not resp.is_success:
@ -150,7 +147,7 @@ class FirecrawlClient:
# "socialMedia": info.get("socialMedia", {}),
"branding": info.get("branding", {}),
"siteLinks": data.get("links", []),
"html": data.get("rawHtml", "") or data.get("html", ""), # rawHtml = 가공 전 원본 — <script> 등 보존.
"html": data.get("html", ""), # raw HTML — collect 단계에서 tracking 추출 후 raw_data 에는 저장 안 함.
"sourceUrl": url,
}
@ -190,7 +187,6 @@ class FirecrawlClient:
},
},
"waitFor": 5000,
"maxAge": 0,
},
timeout=60,
label="firecrawl-gangnamunni",

View File

@ -219,7 +219,7 @@ class VisionClient:
' "logo_symbol": "심볼이 묘사하는 대상 (예: \'잎사귀\', \'추상 곡선\'). 없으면 빈 문자열",\n'
' "logo_text": "로고에 보이는 워드마크 텍스트 그대로 (한글/영문). 없으면 빈 문자열",\n'
' "logo_colors_desc": "로고에 쓰인 색감을 사람이 부르는 이름으로 서술 (예: \'딥네이비 + 골드\')",\n'
' "logo_colors_hex": ["로고에서 시각적으로 가장 두드러진 색 최대 2개의 hex 근사값 배열. 예: [\'#1A2B3C\', \'#D4A017\']. 색이 1개면 1개만, 강한 색이 1개도 없으면 빈 배열."]\n'
' "logo_colors_hex": ["로고에서 시각적으로 두드러진 색 정확히 5개의 hex 근사값 배열. 예: [\'#1A2B3C\', \'#D4A017\', \'#FFFFFF\', \'#9E5C2A\', \'#1F1F1F\']. 강한 색이 5개 안 되면 음영/명도 차이로 5개 채울 것. 빈 배열 금지."]\n'
"}\n"
"주의: logo_colors_hex 는 시각 추정이라 정확도 떨어질 수 있음. CSS 추출이 우선이고 이건 fallback/보완 용.\n"
"모든 설명/텍스트 값은 반드시 한국어로 작성하세요 (영어 금지)."
@ -229,9 +229,14 @@ class VisionClient:
return {}
# logo_images는 우리가 직접 채움 (Vision은 묘사만)
result["logo_images"] = {"circle": None, "horizontal": logo_url, "korean": None}
# logo_colors_hex 최대 2개로 제한.
# logo_colors_hex 5개 강제 정규화 — LLM 이 4개나 6개 줄 수도 있어서 길이 fallback.
hex_list = [h for h in (result.get("logo_colors_hex") or []) if isinstance(h, str) and h.startswith("#")]
result["logo_colors_hex"] = hex_list[:2]
if hex_list:
while len(hex_list) < 5:
hex_list.append(hex_list[-1]) # 마지막 색 복제로 패딩
result["logo_colors_hex"] = hex_list[:5]
else:
result["logo_colors_hex"] = []
return result
async def describe_channel_logos(

View File

@ -13,7 +13,7 @@ from integrations.llm.schemas.report import (
ScoresInput, ScoresOutput,
OtherChannelsInput, OtherChannelsOutput
)
from integrations.llm.schemas.plan import PlanInput, PlanOutput, SummarizeInput, SummarizeOutput
from integrations.llm.schemas.plan import PlanInput, PlanOutput
from integrations.llm.schemas.market import (
MarketCompetitorsInput, MarketCompetitorsOutput,
MarketKeywordsInput, MarketKeywordsOutput,
@ -64,13 +64,6 @@ plan_prompt = Prompt(
output_class=PlanOutput,
)
summarize_prompt = Prompt(
file_name="summarize_prompt.txt",
prompt_model="PLAN_MODEL",
input_class=SummarizeInput,
output_class=SummarizeOutput,
)
market_competitors_prompt = Prompt(
file_name="market_competitors_prompt.txt",
prompt_model="MARKET_MODEL",

View File

@ -2,15 +2,6 @@ from typing import Literal
from pydantic import BaseModel
class SummarizeInput(BaseModel):
label: str
data: str
class SummarizeOutput(BaseModel):
summary: str
class PlanInput(BaseModel):
clinic_name: str | None = None
clinic_name_en: str | None = None

View File

@ -134,7 +134,7 @@ class YouTubeAudit(BaseModel):
channel_description: str
linked_urls: list[LinkedUrl]
playlists: list[str]
top_videos: list[TopVideo] = []
top_videos: list[TopVideo]
diagnosis: list[DiagnosisItem]
@ -184,11 +184,11 @@ class FacebookPage(BaseModel):
followers: int
following: int
category: str
bio: str = ""
bio: str
logo: str
logo_description: str
link: str
linked_domain: str = ""
linked_domain: str
reviews: int
recent_post_age: str
has_whatsapp: bool | None = None
@ -245,7 +245,7 @@ class AdditionalDomain(BaseModel):
class WebsiteAudit(BaseModel):
primary_domain: str
additional_domains: list[AdditionalDomain] = []
additional_domains: list[AdditionalDomain]
sns_links_on_site: bool
sns_links_detail: list[SnsLink] | None = None
tracking_pixels: list[TrackingPixel]

View File

@ -1,7 +0,0 @@
다음은 "{label}" 원본 데이터입니다.
{data}
위 데이터를 마케팅 플랜 생성에 필요한 핵심 정보만 남기고 간결하게 요약하세요.
구체적인 수치·날짜·고유명사(채널명, 게시물 제목 등)는 그대로 보존하고, 중복되거나 플랜 작성에 불필요한 메타데이터는 제거하세요.
요약문 하나의 문자열로만 출력하세요.

View File

@ -48,7 +48,7 @@ class YouTubeClient:
resp = await http_request(
HTTPMethod.GET,
url=f"{YT}/channels",
params={"part": "snippet,statistics,contentDetails", "id": channel_id, "key": self.api_key},
params={"part": "snippet,statistics", "id": channel_id, "key": self.api_key},
label="yt-channel",
)
if not resp or not resp.is_success:
@ -58,52 +58,26 @@ class YouTubeClient:
return None
channel = items[0]
async def _video_details(video_ids: list[str]) -> list[dict]:
"""video_ids 순서를 보존한 채 snippet/statistics/contentDetails 채워서 반환."""
if not video_ids:
return []
video_ids: list[str] = []
resp = await http_request(
HTTPMethod.GET,
url=f"{YT}/search",
params={"part": "snippet", "channelId": channel_id, "order": "viewCount", "type": "video", "maxResults": 10, "key": self.api_key},
label="yt-search",
)
if resp and resp.is_success:
video_ids = [i["id"]["videoId"] for i in resp.json().get("items", []) if i.get("id", {}).get("videoId")]
videos: list[dict] = []
if video_ids:
resp = await http_request(
HTTPMethod.GET,
url=f"{YT}/videos",
params={"part": "snippet,statistics,contentDetails", "id": ",".join(video_ids), "key": self.api_key},
label="yt-videos",
)
if not resp or not resp.is_success:
return []
video_map = {v["id"]: v for v in resp.json().get("items", [])}
return [video_map[vid] for vid in video_ids if vid in video_map]
# 인기 영상 top 10 (조회수순) — search index 기반, 정확한 정렬 보장.
resp = await http_request(
HTTPMethod.GET,
url=f"{YT}/search",
params={"part": "snippet", "channelId": channel_id, "order": "viewCount", "type": "video", "maxResults": 10, "key": self.api_key},
label="yt-search-top",
)
top_ids: list[str] = []
if resp and resp.is_success:
top_ids = [i["id"]["videoId"] for i in resp.json().get("items", []) if i.get("id", {}).get("videoId")]
videos = await _video_details(top_ids)
# 최근 영상 10개 — search index는 누락이 흔해 채널의 실제 uploads 재생목록에서 직접 읽고
# publishedAt 기준으로 코드에서 직접 정렬 (재생목록 순서 자체는 보장되지 않으므로).
recents: list[dict] = []
uploads_id = (channel.get("contentDetails") or {}).get("relatedPlaylists", {}).get("uploads")
if uploads_id:
resp = await http_request(
HTTPMethod.GET,
url=f"{YT}/playlistItems",
params={"part": "snippet", "playlistId": uploads_id, "maxResults": 10, "key": self.api_key},
label="yt-uploads",
)
if resp and resp.is_success:
entries = resp.json().get("items", [])
entries.sort(key=lambda i: i.get("snippet", {}).get("publishedAt") or "", reverse=True)
recent_ids = [
vid for e in entries
if (vid := e.get("snippet", {}).get("resourceId", {}).get("videoId"))
]
recents = await _video_details(recent_ids)
videos = resp.json().get("items", [])[:10]
playlists: list[dict] = []
resp = await http_request(
@ -115,19 +89,7 @@ class YouTubeClient:
if resp and resp.is_success:
playlists = resp.json().get("items", [])
return {"channelId": channel_id, "channel": channel, "videos": videos, "recents": recents, "playlists": playlists}
@staticmethod
def _video_item(v: dict) -> dict:
return {
"title": v.get("snippet", {}).get("title"),
"views": int(v.get("statistics", {}).get("viewCount", 0)),
"likes": int(v.get("statistics", {}).get("likeCount", 0)),
"comments": int(v.get("statistics", {}).get("commentCount", 0)),
"date": v.get("snippet", {}).get("publishedAt"),
"duration": v.get("contentDetails", {}).get("duration"),
"url": f"https://www.youtube.com/watch?v={v['id']}",
}
return {"channelId": channel_id, "channel": channel, "videos": videos, "playlists": playlists}
async def get_channel(self, url: str) -> dict | None:
raw = await self.fetch_channel(url)
@ -148,8 +110,18 @@ class YouTubeClient:
"subscribers": int(stats.get("subscriberCount", 0)),
"totalViews": int(stats.get("viewCount", 0)),
"totalVideos": int(stats.get("videoCount", 0)),
"videos": [self._video_item(v) for v in raw["videos"]],
"recents": [self._video_item(v) for v in raw["recents"]],
"videos": [
{
"title": v.get("snippet", {}).get("title"),
"views": int(v.get("statistics", {}).get("viewCount", 0)),
"likes": int(v.get("statistics", {}).get("likeCount", 0)),
"comments": int(v.get("statistics", {}).get("commentCount", 0)),
"date": v.get("snippet", {}).get("publishedAt"),
"duration": v.get("contentDetails", {}).get("duration"),
"url": f"https://www.youtube.com/watch?v={v['id']}",
}
for v in raw["videos"]
],
"playlists": [
p.get("snippet", {}).get("title")
for p in raw["playlists"]

View File

@ -130,7 +130,7 @@ class YouTubeAudit(CamelModel):
channel_description: str
linked_urls: list[LinkedUrl]
playlists: list[str]
top_videos: list[TopVideo] = []
top_videos: list[TopVideo]
diagnosis: list[DiagnosisItem]
@ -176,11 +176,11 @@ class FacebookPage(CamelModel):
followers: int
following: int
category: str
bio: str = ""
bio: str
logo: str
logo_description: str
link: str
linked_domain: str = ""
linked_domain: str
reviews: int
recent_post_age: str
has_whatsapp: bool | None = None
@ -222,7 +222,7 @@ class AdditionalDomain(CamelModel):
class WebsiteAudit(CamelModel):
primary_domain: str
additional_domains: list[AdditionalDomain] = []
additional_domains: list[AdditionalDomain]
sns_links_on_site: bool
sns_links_detail: list[SnsLink] | None = None
tracking_pixels: list[TrackingPixel]

View File

@ -1,4 +1,3 @@
import asyncio
import json
import logging
from urllib.parse import urlparse
@ -8,7 +7,7 @@ from common.db.run import update_run_report, update_run_plan, select_run_report_
from common.db.source import select_run_raw_data, select_mainpage_logo_url
from common.db.market import select_market
from integrations.llm.llm_service import LLMService
from integrations.llm.prompt import report_prompt, plan_prompt, summarize_prompt, youtube_diagnosis_prompt, brand_consistency_prompt, critical_issues_prompt, transformation_prompt, roadmap_prompt, scores_prompt, other_channels_prompt
from integrations.llm.prompt import report_prompt, plan_prompt, youtube_diagnosis_prompt, brand_consistency_prompt, critical_issues_prompt, transformation_prompt, roadmap_prompt, scores_prompt, other_channels_prompt
from integrations.llm.schemas.report import ReportOutput, ClinicSnapshot, YouTubeAudit, BrandConsistencyOutput, CriticalIssuesOutput, DiagnosisItem, TransformationProposal, RoadmapOutput, RoadmapMonth, ScoresOutput, ChannelScore, WebsiteAudit, OtherChannelsOutput, OtherChannel
from services.branding import analyze_branding
from services.instagram_audit import build_instagram_audit
@ -50,26 +49,6 @@ async def generate_plan(analysis_run_id: str) -> PlanOutput:
def _json(v) -> str | None:
return json.dumps(v, ensure_ascii=False) if v else None
# map: 큰 입력은 LLM으로 압축 요약해 100KB 초과 에러를 방지하고, 작은 입력은 그대로 둔다.
large_fields = {
"report": _json(report),
"market_competitors": _json(market.get("competitors")),
"market_keywords": _json(market.get("keywords")),
"market_trend": _json(market.get("trend")),
"market_target_audience": _json(market.get("target_audience")),
"tiktok": _json(tiktok),
"instagram": _json(instagram),
"facebook": _json(facebook),
"naver_cafe": _json(naver_cafe),
"channel_logos": _json(branding.get("channelLogos")),
"brand_assets": _json(branding.get("brandAssets")),
}
summarized = dict(zip(
large_fields.keys(),
await asyncio.gather(*(_summarize(label, data) for label, data in large_fields.items())),
))
# reduce: 요약된 입력을 모아 최종 플랜 생성.
input_data = {
"clinic_name": clinic.get("clinicName"),
"clinic_name_en": clinic.get("clinicNameEn"),
@ -78,24 +57,24 @@ async def generate_plan(analysis_run_id: str) -> PlanOutput:
"slogan": clinic.get("slogan"),
"services": json.dumps(clinic.get("services", []), ensure_ascii=False),
"doctors": json.dumps(clinic.get("doctors", []), ensure_ascii=False),
"report": _json(report),
"market_competitors": _json(market.get("competitors")),
"market_keywords": _json(market.get("keywords")),
"market_trend": _json(market.get("trend")),
"market_target_audience": _json(market.get("target_audience")),
"tiktok": _json(tiktok),
"instagram": _json(instagram),
"facebook": _json(facebook),
"naver_blog": _json(_naver_blog_summary(naver_blog)),
"naver_cafe": _json(naver_cafe),
"kakao_talk": _json(kakaotalk),
**summarized,
"channel_logos": _json(branding.get("channelLogos")),
"brand_assets": _json(branding.get("brandAssets")),
}
return await LLMService(provider="perplexity").generate(plan_prompt, input_data)
_SUMMARIZE_THRESHOLD = 4000 # 이 길이(문자 수)를 넘는 입력만 요약 LLM 호출 (불필요한 호출 방지)
async def _summarize(label: str, data: str | None) -> str | None:
if not data or len(data) <= _SUMMARIZE_THRESHOLD:
return data
result = await LLMService(provider="perplexity").generate(summarize_prompt, {"label": label, "data": data})
return result.summary
def _build_clinic_snapshot(mainpage: dict, gangnam_unni: dict, brand_assets: dict, logo_url: str | None) -> dict:
snapshot: dict = {}
doctors = gangnam_unni.get("doctors", [])
@ -135,13 +114,12 @@ def _naver_blog_summary(blog: dict | None) -> dict | None:
async def _build_youtube_audit(youtube: dict) -> dict: # 기획상 1개의 input channel, 다중 채널은 기획에 없음.
videos = youtube.get("videos", [])
recents = youtube.get("recents", [])
yt_patch: dict = {
"weekly_view_growth": {"absolute": 0, "percentage": 0.0},
"estimated_monthly_revenue": {"min": 0, "max": 0},
"linked_urls": [],
"avg_video_length": calc_avg_video_length(videos),
"upload_frequency": calc_upload_frequency(recents),
"upload_frequency": calc_upload_frequency(videos),
}
if youtube.get("channelName"): yt_patch["channel_name"] = youtube["channelName"]
if youtube.get("handle"): yt_patch["handle"] = youtube["handle"]
@ -363,9 +341,8 @@ async def run_plan_task(analysis_run_id: str) -> None:
result = await generate_plan(analysis_run_id)
# profile_photo 는 brand_assets.logo_description 으로 코드가 박음 (LLM "(가이드 미보유)" 같은 hallucination 차단).
raw = await select_run_raw_data(analysis_run_id)
branding_list = raw.get(SourceType.BRANDING) or []
branding_data = branding_list[0]["raw_data"] if branding_list else {}
logo_desc = (branding_data.get("brandAssets") or {}).get("logo_description") or ""
branding = raw.get(SourceType.BRANDING) or []
logo_desc = (((branding[0] if branding else {}).get("brandAssets") or {}).get("logo_description")) or ""
result = _patch_plan(result, logo_desc)
await update_run_plan(analysis_run_id, result.model_dump())
logger.info("[plan] done run=%s", analysis_run_id)

View File

@ -39,11 +39,7 @@ async def _describe_logo(analysis_run_id: str, info_id: int, vc: VisionClient) -
if result:
break
if result:
# collect_brand_basics 가 미리 채운 brand_colors/color_palette/color_source 보존하면서 logo_* 덧붙이기.
raw = await select_run_raw_data(analysis_run_id)
existing = ((raw.get("branding") or [{}])[0].get("raw_data") or {}).get("brandAssets") or {}
merged = {**existing, **result}
await update_raw_info_merge(info_id, {"brandAssets": merged})
await update_raw_info_merge(info_id, {"brandAssets": result})
logger.info("[brand_logo] done keys=%s", list(result.keys()) if result else None)

View File

@ -86,15 +86,15 @@ async def collect_gangnam_unni(analysis_run_id: str, info_id: int, url: str) ->
logger.info("[gangnam_unni] done run=%s", analysis_run_id)
def _extract_website_audit(html: str, url: str) -> dict:
"""raw HTML 에서 main CTA / tracking pixels / SNS / additional domains 정규식 추출."""
def _extract_website_audit(html: str, site_links: list[str], url: str) -> dict:
"""raw HTML + siteLinks 에서 main CTA / tracking pixels / SNS / additional domains 정규식 추출."""
if not html:
return {}
primary_host = urlparse(url).netloc
result: dict = {
"trackingPixels": extract_tracking_pixels(html),
"snsLinks": extract_sns_links(html),
"additionalDomains": extract_additional_domains(html, primary_host),
"snsLinks": extract_sns_links(site_links, html),
"additionalDomains": extract_additional_domains(site_links, primary_host),
}
html_cta = extract_main_cta(html)
if html_cta:
@ -115,7 +115,7 @@ async def collect_mainpage(analysis_run_id: str, info_id: int, hospital_id: str,
html = data.pop("html", "") or "" # raw_data 에는 저장 안 함 — 여기서만 사용하고 버림.
# 홈페이지 URL 자체도 raw_data 에 박아둬야 brand_assets / 분석 단계에서 mainpage URL 재조회 없이 사용 가능.
data = {**data, "sourceUrl": url}
data.update(_extract_website_audit(html, url))
data.update(_extract_website_audit(html, data.get("siteLinks", []), url))
await update_raw_info(info_id, data)
await update_hospital(hospital_id, data, analysis_run_id=analysis_run_id)

View File

@ -89,8 +89,7 @@ def _logo_data(channel_logos: dict, channel: str) -> dict:
}
def _page_patch(item: dict, channel_logos) -> dict:
p: dict = {"page_name": "", "followers": 0, "category": "", "reviews": 0,
"following": 0, "recent_post_age": "", "post_frequency": "", "engagement": ""}
p: dict = {}
fb = item["raw_data"]
language = item.get("language") if item.get("language") else "KR"
label = "페이스북 " + language

View File

@ -64,8 +64,8 @@ def build_kpi_dashboard(
print("facebook_kpis", facebook_kpis)
print("instagram_kpis", instagram_kpis)
kpis += [k for k in facebook_kpis if k]
kpis += [k for k in instagram_kpis if k]
kpis += facebook_kpis
kpis += instagram_kpis
for k in [
_follower_kpi("YouTube 구독자", youtube.get("subscribers")),
_follower_kpi("TikTok 팔로워", tiktok.get("followers")),

View File

@ -1,11 +1,13 @@
"""HTML 정규식 기반 deterministic 추출 — tracking pixels / SNS / additional domains / main CTA."""
"""collect 단계 - HTML / siteLinks 에서 tracking pixels / SNS links / additional domains 추출.
모두 정규식·도메인 매칭 기반 deterministic 추출 (LLM 미경유)."""
import re
from urllib.parse import urlparse
# ── tracking pixels ──────────────────────────────────────────────────────────
# 픽셀별 시그니처: 하나라도 매치되면 installed, ID 는 첫 non-empty 그룹.
# 픽셀별 시그니처: 이름 → 정규식 리스트. installed 판정은 OR (하나라도 매치되면 설치된 것).
# ID 캡처는 모든 패턴 시도 후 첫 non-empty 그룹 사용 (signature-only 패턴과 ID-capture 패턴 혼재 OK).
_TRACKING_PIXEL_PATTERNS: dict[str, list[re.Pattern]] = {
"Google Analytics": [
re.compile(r"googletagmanager\.com/gtag/js\?id=(G-[A-Z0-9]+)", re.IGNORECASE),
@ -37,7 +39,8 @@ _TRACKING_PIXEL_PATTERNS: dict[str, list[re.Pattern]] = {
def extract_tracking_pixels(html: str) -> list[dict]:
"""HTML 에서 트래킹 픽셀 설치 여부 + ID 추출."""
"""HTML 에서 트래킹 픽셀 설치 여부 + ID 추출. 환각 0 — 정규식 매치만 신뢰.
모든 패턴 시도 하나라도 매치되면 installed. ID 캡처된 non-empty 그룹 사용."""
if not html:
return []
pixels: list[dict] = []
@ -63,15 +66,9 @@ def extract_tracking_pixels(html: str) -> list[dict]:
# ── main CTA ─────────────────────────────────────────────────────────────────
_CTA_KEYWORDS = re.compile(
r"(상담|예약|문의|신청|등록|Book\s*Now|Consult|Reservation|Contact|Apply)",
r"(상담|예약|문의|신청|등록|진료|Book\s*Now|Consult|Reservation|Contact|Apply)",
re.IGNORECASE,
)
# CTA 카테고리 우선순위 (전화>카톡>예약). 매치 없으면 결과에 포함 X.
_CTA_CATEGORIES: list[tuple[str, re.Pattern]] = [
("전화 상담", re.compile(r"전화|\bCall\b|\bPhone\b", re.IGNORECASE)),
("카카오톡 상담", re.compile(r"카(카오)?톡|Kakao", re.IGNORECASE)),
("온라인 예약", re.compile(r"예약|Reserv|Book", re.IGNORECASE)),
]
_BUTTON_TAG = re.compile(r"<button\b[^>]*>(.*?)</button>", re.IGNORECASE | re.DOTALL)
_ANCHOR_TAG = re.compile(r"<a\b[^>]*>(.*?)</a>", re.IGNORECASE | re.DOTALL)
@ -81,37 +78,22 @@ def _clean_text(inner: str) -> str:
return re.sub(r"\s+", " ", text).strip()
_MAIN_CTA_LIMIT = 3
def _classify_cta(text: str) -> str | None:
"""CTA 텍스트 → 카테고리 (전화/카톡/예약). 매치 없으면 None."""
for label, pat in _CTA_CATEGORIES:
if pat.search(text):
return label
return None
def extract_main_cta(html: str) -> str:
"""<button>·<a> 라벨을 카테고리로 분류 후 우선순위 순서로 최대 _MAIN_CTA_LIMIT 개 join."""
"""HTML 에서 primary CTA 텍스트 추출. 우선순위: <button> 첫 매치 → <a> 첫 매치.
CTA 키워드 매칭 + 길이 1~20 (버튼 라벨 추정)."""
if not html:
return ""
found: set[str] = set()
for pat in (_BUTTON_TAG, _ANCHOR_TAG):
for m in pat.finditer(html):
text = _clean_text(m.group(1))
if text and 1 <= len(text) <= 20 and _CTA_KEYWORDS.search(text):
cat = _classify_cta(text)
if cat:
found.add(cat)
ordered_labels = [label for label, _ in _CTA_CATEGORIES]
result = [label for label in ordered_labels if label in found]
return " + ".join(result[:_MAIN_CTA_LIMIT])
return text
return ""
# ── SNS links ────────────────────────────────────────────────────────────────
# SNS 도메인 → 표준 platform 이름.
# SNS 도메인 → 표준 platform 이름. siteLinks 필터링 + DOM 위치 판정에 공유.
_SNS_DOMAINS: dict[str, str] = {
"facebook.com": "Facebook",
"instagram.com": "Instagram",
@ -128,16 +110,11 @@ _SNS_DOMAINS: dict[str, str] = {
_FOOTER_BLOCK = re.compile(r"<footer\b[^>]*>(.*?)</footer>", re.IGNORECASE | re.DOTALL)
_HEADER_BLOCK = re.compile(r"<header\b[^>]*>(.*?)</header>", re.IGNORECASE | re.DOTALL)
# <a> 태그 안의 href 만 매치 — <link>/<script> 등 리소스 호출 제외 (CDN 노이즈 차단).
_ANCHOR_HREF_PATTERN = re.compile(
r"""<a\b[^>]*\bhref\s*=\s*['"](https?://[^'"\s]+)['"]""",
re.IGNORECASE,
)
def _platform_for(url: str) -> str | None:
try:
host = urlparse(url).netloc.lower().removeprefix("www.")
host = urlparse(url).netloc.lower().lstrip("www.")
except Exception:
return None
for domain, name in _SNS_DOMAINS.items():
@ -147,7 +124,7 @@ def _platform_for(url: str) -> str | None:
def _location_for(url: str, html: str) -> str:
"""url 이 <header>/<footer> 블록 안에 있는지 판정 — 둘 다 아니면 'body'."""
"""url 이 HTML 의 <header>/<footer> 안에 들어있는지로 위치 판정. 둘 다 아니면 'body'."""
if not html:
return ""
needle = re.escape(url)
@ -160,21 +137,11 @@ def _location_for(url: str, html: str) -> str:
return "body"
def _extract_anchor_hrefs(html: str) -> list[str]:
"""<a href="http..."> 만 추출 — 순서 보존, 중복 제거."""
if not html:
return []
seen: dict[str, None] = {}
for url in _ANCHOR_HREF_PATTERN.findall(html):
if url not in seen:
seen[url] = None
return list(seen.keys())
def extract_sns_links(html: str) -> list[dict]:
"""anchor href 에서 SNS 도메인 매칭. 중복 platform 은 첫 URL 만 유지."""
def extract_sns_links(site_links: list[str], html: str = "") -> list[dict]:
"""siteLinks 에서 SNS 도메인 매칭. location 은 HTML 의 footer/header tag 블록으로 판정.
중복 platform URL 유지."""
seen: dict[str, dict] = {}
for url in _extract_anchor_hrefs(html):
for url in site_links or []:
platform = _platform_for(url)
if not platform or platform in seen:
continue
@ -186,63 +153,26 @@ def extract_sns_links(html: str) -> list[dict]:
return list(seen.values())
# ── additional domains (글로벌/다국어 사이트만) ─────────────────────────────────
# ── additional domains ───────────────────────────────────────────────────────
# 언어 코드 → 한국어 라벨.
_LANG_LABEL: dict[str, str] = {
"en": "영어", "eng": "영어",
"zh": "중국어", "cn": "중국어", "chn": "중국어",
"ja": "일본어", "jp": "일본어",
"ko": "한국어", "kor": "한국어", "kr": "한국어",
"vi": "베트남어", "vn": "베트남어",
"th": "태국어", "thai": "태국어",
"ru": "러시아어",
"es": "스페인어",
"mn": "몽골어",
"ar": "아랍어", "arab": "아랍어",
"id": "인도네시아어",
"de": "독일어",
"fr": "프랑스어",
"pt": "포르투갈어",
def extract_additional_domains(site_links: list[str], primary_host: str) -> list[dict]:
"""siteLinks 의 host 중 primary 도메인이 아닌 외부 도메인 모으기. SNS·맵 등 utility 도메인 제외.
purpose 분류는 LLM 필요 여기선 문자열."""
skip_domains = set(_SNS_DOMAINS) | {
"map.kakao.com", "map.naver.com", "maps.google.com", "goo.gl",
}
_LANG_LI_PATTERN = re.compile(
r'<li\b[^>]*\bdata-lang\s*=\s*[\'"]([^\'"]+)[\'"][^>]*>(.*?)</li>',
re.IGNORECASE | re.DOTALL,
)
def extract_additional_domains(html: str, primary_host: str) -> list[dict]:
"""글로벌/다국어 사이트 anchor 수집 — data-lang 마커 + URL 서브도메인 prefix 로 식별."""
if not html:
return []
primary = (primary_host or "").lower().removeprefix("www.")
primary = (primary_host or "").lower().lstrip("www.")
seen: dict[str, dict] = {}
def add(host: str, lang_kr: str):
host = host.lower().removeprefix("www.")
if not host or host == primary or host in seen:
return
seen[host] = {"domain": host, "purpose": lang_kr}
# 1) <li data-lang="xx"> 안의 anchor
for m in _LANG_LI_PATTERN.finditer(html):
lang_kr = _LANG_LABEL.get(m.group(1).lower())
if not lang_kr:
for url in site_links or []:
if not url or not url.startswith("http"):
continue
a_m = re.search(r'<a[^>]*href\s*=\s*[\'"](https?://[^\'"\s]+)[\'"]', m.group(2), re.IGNORECASE)
if a_m:
add(urlparse(a_m.group(1)).netloc, lang_kr)
# 2) URL 서브도메인 prefix 가 언어 코드
for url in _extract_anchor_hrefs(html):
host = urlparse(url).netloc.lower().removeprefix("www.")
parts = host.split(".")
if len(parts) >= 3:
lang_kr = _LANG_LABEL.get(parts[0])
if lang_kr:
add(host, lang_kr)
try:
host = urlparse(url).netloc.lower().lstrip("www.")
except Exception:
continue
if not host or host == primary or host in seen:
continue
if any(host == d or host.endswith("." + d) for d in skip_domains):
continue
seen[host] = {"domain": host, "purpose": ""}
return list(seen.values())

View File

@ -12,8 +12,3 @@ services:
volumes:
- ./app:/app
restart: unless-stopped
networks:
- o2o-net
networks:
o2o-net:
external: true