DB_JOB/작업파일/_검수_N대조.py

# -*- coding: utf-8 -*-
"""사이트 직접 대조 검수 — N(저작물 유형)·L 이상치 탐지(읽기전용 audit).
각 데이터행 페이지/게시판을 실제 렌더해 본문 콘텐츠 이미지/영상을 '원본크기' 기준 탐지하고
시트 N과 비교해 과표시/누락/모순 이상치만 보고 + 의심행 실제이미지 몽타주.

⚠ 읽기전용: 엑셀을 수정하지 않는다. 사람이 몽타주 보고 최종 판정.
⚠ 콘텐츠 이미지 경로는 사이트마다 다름(kesco=/common/images/sub) → 경로(/common)로 거르지 말고
   파일명 기준 데코필터(logo/icon/btn/mark_…) + naturalWidth 사용. (kesco 교훈)

사용:
  python -X utf8 _검수_N대조.py <엑셀경로> [--rows 67-끝|3-끝] [--limit N]
출력: <기관>_naudit\ (의심행 실이미지 + montage_*.png) + naudit.json + 콘솔 이상치표
"""
import sys, os, re, json, io, warnings, argparse
import openpyxl, requests
from PIL import Image, ImageDraw, ImageFont
warnings.filterwarnings('ignore'); requests.packages.urllib3.disable_warnings()
UA = 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/120.0.0.0 Safari/537.36'
# 파일명 기준 데코(절대 콘텐츠 아님). 경로(/common 등)로 거르지 말 것.
NAME_NOISE = re.compile(r'(^ico[_\-]|icon|logo|btn[_\-]|^bul|bullet|^bg[_\-]|banner|^sns|blank|spacer|loading|arrow|^dot|^line[_\.]|^top_|footer|^foot|^header|symbol|copyright|^qr|no_img|noimage|share|facebook|insta|twitter|naver|kakao|mark_|webwatch|eprivacy|wa\.png)', re.I)
PATH_NOISE = re.compile(r'(/skin/|/template/|/editor/|/sample)', re.I)
CONT_SEL = '#contents,.contents,#content,.content,.sub_content,.sub_contents,#sub_content,#contents_area,.contents_area,#container,#contentBody,.board_view,.view_con'
JS = r"""(sel)=>{
  const cont=document.querySelector(sel)||document.body;
  const imgs=[];
  for(const i of cont.querySelectorAll('img')){imgs.push({src:i.currentSrc||i.src||'',alt:(i.alt||''),nw:i.naturalWidth,nh:i.naturalHeight});}
  let vid=0;
  if(cont.querySelector('video,source[src*=".mp4"]'))vid++;
  for(const f of cont.querySelectorAll('iframe')){const s=f.src||'';if(/youtube|youtu\.be|vimeo|player\./i.test(s))vid++;}
  return {imgs,vid};
}"""


def real_imgs(items):
    out, seen = [], set()
    for it in items:
        src = it.get('src', '')
        if not src or src in seen:
            continue
        base = src.split('?')[0].split('/')[-1]
        if NAME_NOISE.search(base) or PATH_NOISE.search(src):
            continue
        nw, nh = it.get('nw', 0), it.get('nh', 0)
        if nw >= 200 and nh >= 150 and nw * nh >= 40000:
            seen.add(src); out.append(it)
    return out


def parse_rows(spec, lo, hi):
    if not spec:
        return lo, hi
    a, b = spec.split('-')
    a = int(a); b = hi if b in ('끝', 'end', '') else int(b)
    return a, b


def run():
    ap = argparse.ArgumentParser()
    ap.add_argument('xlsx'); ap.add_argument('--rows', default='')
    ap.add_argument('--limit', type=int, default=0)
    a = ap.parse_args()
    wb = openpyxl.load_workbook(a.xlsx); ws = wb.active
    name = os.path.splitext(os.path.basename(a.xlsx))[0]
    OUT = os.path.join(os.path.dirname(a.xlsx), f'{name}_naudit'); os.makedirs(OUT, exist_ok=True)
    r0, r1 = parse_rows(a.rows, 3, ws.max_row)
    targets = []
    for r in range(r0, r1 + 1):
        if ws.cell(r, 2).value is None and all(ws.cell(r, c).value in (None, '') for c in range(4, 12)):
            continue
        L = ws.cell(r, 12).value or ''
        M = ws.cell(r, 13).value
        N = ws.cell(r, 14).value or ''
        url = ws.cell(r, 11).value
        targets.append((r, url, L, M, N))
    if a.limit:
        targets = targets[:a.limit]
    print(f'[{name}] 대조 대상 {len(targets)}행', flush=True)
    from playwright.sync_api import sync_playwright
    plan = []; dl = []
    sess = requests.Session(); sess.headers['User-Agent'] = UA
    with sync_playwright() as p:
        b = p.chromium.launch()
        pg = b.new_page(user_agent=UA, viewport={'width': 1280, 'height': 2000})
        for idx, (r, url, L, M, N) in enumerate(targets):
            rec = {'row': r, 'L': L, 'M': M, 'N': N, 'url': url}
            # 사이트면 뒤컬럼 공백이어야(11항)
            if L == '사이트':
                if N or (M not in (None, '')):
                    rec['flag'] = 'SITE_TAIL'; rec['note'] = f'사이트인데 M={M} N={N}'
                plan.append(rec); continue
            if not (url and isinstance(url, str) and url.startswith('http')):
                plan.append(rec); continue
            try:
                pg.goto(url, timeout=22000, wait_until='domcontentloaded'); pg.wait_for_timeout(1200)
                try:
                    pg.evaluate("()=>window.scrollTo(0,document.body.scrollHeight)"); pg.wait_for_timeout(800)
                    pg.evaluate("()=>window.scrollTo(0,0)"); pg.wait_for_timeout(200)
                except Exception:
                    pass
                d = pg.evaluate(JS, CONT_SEL)
                ci = real_imgs(d['imgs']); det_img = len(ci); det_vid = d['vid']
                rec['det_img'] = det_img; rec['det_vid'] = det_vid
                has_img = '이미지' in N; has_vid = '영상' in N
                # 이상치 규칙
                if det_img >= 1 and not has_img:
                    rec['flag'] = 'UNDER_IMG'; rec['note'] = f'본문이미지 {det_img}개 검출인데 N에 이미지 없음'
                elif has_img and det_img == 0:
                    rec['flag'] = 'OVER_IMG'; rec['note'] = 'N=이미지인데 본문 실이미지 0(차트/도식/마크 의심)'
                if det_vid and not has_vid:
                    rec['flag'] = (rec.get('flag', '') + '+VID').strip('+'); rec['note'] = (rec.get('note', '') + ' | 영상검출인데 N영상없음').strip()
                if L == '게시판' and (M == 0) and N != '없음':
                    rec['flag'] = (rec.get('flag', '') + '+BBS0').strip('+'); rec['note'] = (rec.get('note', '') + ' | 빈게시판M0인데 N≠없음').strip()
                # 의심행 대표이미지 다운로드(몽타주용)
                if rec.get('flag', '').startswith(('UNDER', 'OVER')):
                    for it in ci[:3]:
                        try:
                            rr = sess.get(it['src'], timeout=10, verify=False)
                            im = Image.open(io.BytesIO(rr.content)).convert('RGB'); im.thumbnail((300, 300))
                            lp = os.path.join(OUT, f'r{r}_{len(dl)}.jpg'); im.save(lp, quality=80)
                            lab = ws.cell(r, 6).value or ws.cell(r, 5).value or ws.cell(r, 4).value or ''
                            dl.append((r, rec['flag'], str(lab)[:16], lp))
                        except Exception:
                            pass
            except Exception as e:
                rec['err'] = str(e)[:40]
            plan.append(rec)
            if (idx + 1) % 25 == 0:
                print(f'  {idx+1}/{len(targets)}', flush=True)
        b.close()
    # 몽타주(의심행만)
    try: font = ImageFont.truetype('malgun.ttf', 15)
    except Exception: font = ImageFont.load_default()
    cols, cw, ch = 5, 240, 270; per = cols * 6; mps = []
    for gi in range(0, len(dl), per):
        chunk = dl[gi:gi + per]; rn = (len(chunk) + cols - 1) // cols
        cv = Image.new('RGB', (cols * cw, rn * ch), 'white'); dr = ImageDraw.Draw(cv)
        for j, (r, flag, lab, lp) in enumerate(chunk):
            try:
                im = Image.open(lp).convert('RGB'); im.thumbnail((cw - 8, ch - 30))
            except Exception: continue
            cx, cy = (j % cols) * cw, (j // cols) * ch; cv.paste(im, (cx + 4, cy + 26))
            dr.rectangle([cx, cy, cx + cw - 1, cy + ch - 1], outline='gray')
            dr.text((cx + 4, cy + 4), f'r{r} {flag}', fill='red', font=font)
        mp = os.path.join(OUT, f'montage_{gi//per+1}.png'); cv.save(mp); mps.append(mp)
    json.dump(plan, open(os.path.join(OUT, 'naudit.json'), 'w', encoding='utf-8'), ensure_ascii=False, indent=1)
    flags = [x for x in plan if x.get('flag')]
    print(f'\n=== 이상치 {len(flags)}건 ===', flush=True)
    for x in flags:
        print(f"  r{x['row']} [{x['flag']}] L={x['L']} N={x['N']} :: {x.get('note','')}", flush=True)
    print(f'\n몽타주 {len(mps)}장:', flush=True)
    for mp in mps: print('  MONT:', mp, flush=True)
    print('  JSON:', os.path.join(OUT, 'naudit.json'), flush=True)


if __name__ == '__main__':
    run()