o2o-infinith-demo/scripts/export_qb_excel.py

#!/usr/bin/env python3
"""Question Bank 120문항 + 2엔진 실측 결과를 엑셀(xlsx)로 내보낸다.

시트: 요약 / 통합결과 / 병원빈도 / GPT상세 / PPLX상세
"""
import json
import os
import time

from openpyxl import Workbook
from openpyxl.styles import Alignment, Font, PatternFill
from openpyxl.utils import get_column_letter

ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
QUESTIONS = os.path.join(ROOT, "scripts", "question_bank_viewclinic.json")
OUT = os.path.join(ROOT, "docs", "reports", "Viewclinic_QB_Results.xlsx")

INK = "0A1128"
VIOLET = "4F1DA1"
LIGHT = "F7F8FC"
GOOD = "DCFCE7"
BAD = "FEE2E2"
WARN = "FEF9C3"

CAT_LABEL = {
    "A": "브랜드 직접", "B": "위치·운영", "C": "지역 추천·비교",
    "D1": "가슴성형", "D2": "눈성형", "D3": "코성형", "D4": "안면윤곽·양악",
    "D5": "지방성형", "D6": "피부·리프팅", "D7": "재수술",
    "E": "신뢰·안전", "F": "가격·이벤트", "G": "후기·평판",
}


def load(engine):
    p = os.path.join(ROOT, "scripts", "out", f"qb_{engine}_results.jsonl")
    rows = {}
    for line in open(p):
        if line.strip():
            r = json.loads(line)
            rows[r["id"]] = r
    return rows


def style_header(ws, ncols, row=1):
    for c in range(1, ncols + 1):
        cell = ws.cell(row=row, column=c)
        cell.font = Font(bold=True, color="FFFFFF", size=11)
        cell.fill = PatternFill("solid", fgColor=INK)
        cell.alignment = Alignment(vertical="center", wrap_text=True)
    ws.freeze_panes = ws.cell(row=row + 1, column=1)


def main():
    qs = json.load(open(QUESTIONS))
    oa, px = load("openai"), load("perplexity")
    wb = Workbook()

    # ── 요약 ─────────────────────────────────────────────
    ws = wb.active
    ws.title = "요약"
    ws["A1"] = "뷰성형외과 AI Discovery Question Bank 실측 결과"
    ws["A1"].font = Font(bold=True, size=16, color=INK)
    ws["A2"] = f"측정일 {time.strftime('%Y-%m-%d')} · OpenAI Responses API + web_search(gpt-4o, 서울 고정) / Perplexity API(sonar)"
    ws["A3"] = "답변엔진 소비자 UI 자동 조회는 약관 위반이므로 정식 API로 측정. 문의 o2oteam@o2o.kr"
    ws["A2"].font = ws["A3"].font = Font(size=10, color="666B7A")

    def brand(q):
        return "뷰성형외과" in q["question"]

    ws.append([])
    ws.append(["구분", "문항 수", "ChatGPT 뷰 언급", "ChatGPT 언급률", "Perplexity 뷰 언급", "Perplexity 언급률"])
    hr = ws.max_row
    for label, group in (("브랜드 질문 (질문에 병원명 포함)", [q for q in qs if brand(q)]),
                         ("논브랜드 질문 (추천·비교·시술 정보)", [q for q in qs if not brand(q)]),
                         ("전체", qs)):
        om = sum(1 for q in group if oa[q["id"]].get("view_mentioned"))
        pm = sum(1 for q in group if px[q["id"]].get("view_mentioned"))
        ws.append([label, len(group), om, f"{om*100//len(group)}%", pm, f"{pm*100//len(group)}%"])
    style_header(ws, 6, hr)
    ws.append([])
    ws.append(["사전판정 분포", "◯ 인용 기대 5", "△ 불안정 35", "✕ 인용 불가 80"])
    ws.cell(row=ws.max_row, column=1).font = Font(bold=True)
    for col, w in zip("ABCDEF", [34, 10, 16, 14, 16, 14]):
        ws.column_dimensions[col].width = w

    # ── 통합결과 ─────────────────────────────────────────
    ws = wb.create_sheet("통합결과")
    ws.append(["ID", "카테고리", "질문", "퍼널", "브랜드", "사전판정",
               "GPT 언급", "GPT 순위", "GPT 상위 병원 (등장순)",
               "PPLX 언급", "PPLX 순위", "PPLX 상위 병원 (등장순)"])
    style_header(ws, 12)
    for q in qs:
        o, p = oa[q["id"]], px[q["id"]]
        oc = ", ".join([c for c in o.get("clinics", []) if isinstance(c, str)][:5])
        pc = ", ".join([c for c in p.get("clinics", []) if isinstance(c, str)][:5])
        ws.append([q["id"], CAT_LABEL[q["id"].split("-")[0]], q["question"], q["stage"],
                   "O" if brand(q) else "", q["prior"],
                   "O" if o.get("view_mentioned") else "X", o.get("view_rank") or "",
                   oc or "(병원 언급 없음)",
                   "O" if p.get("view_mentioned") else "X", p.get("view_rank") or "",
                   pc or "(병원 언급 없음)"])
        r = ws.max_row
        for col, eng in ((7, o), (10, p)):
            ws.cell(row=r, column=col).fill = PatternFill(
                "solid", fgColor=GOOD if eng.get("view_mentioned") else BAD)
        for c in range(1, 13):
            ws.cell(row=r, column=c).alignment = Alignment(vertical="top", wrap_text=c in (3, 9, 12))
    for col, w in zip(range(1, 13), [8, 13, 44, 7, 7, 9, 9, 9, 44, 9, 9, 44]):
        ws.column_dimensions[get_column_letter(col)].width = w

    # ── 병원빈도 (논브랜드 한정) ─────────────────────────
    ws = wb.create_sheet("병원빈도")
    ws.append(["엔진", "병원", "언급 질문 수", "1순위 등장 수"])
    style_header(ws, 4)

    def canon(name):
        n = name.strip().lower().replace(" ", "")
        if ("view" in n and ("plastic" in n or "성형" in n)) or n.startswith("뷰성형"):
            return "뷰성형외과"
        for suf in ("성형외과의원", "성형외과", "의원", "클리닉", "clinic", "plasticsurgery"):
            if n.endswith(suf) and len(n) > len(suf):
                return n[: -len(suf)]
        return n

    for eng_label, data in (("ChatGPT(OpenAI)", oa), ("Perplexity", px)):
        freq, label = {}, {}
        for q in qs:
            if brand(q):
                continue
            seen = set()
            for i, c in enumerate([c for c in data[q["id"]].get("clinics", []) if isinstance(c, str)][:5]):
                key = canon(c)
                if key in seen:
                    continue
                seen.add(key)
                label.setdefault(key, c)
                freq.setdefault(key, [0, 0])
                freq[key][0] += 1
                if i == 0:
                    freq[key][1] += 1
        for key, (cnt, first) in sorted(freq.items(), key=lambda x: -x[1][0]):
            ws.append([eng_label, label[key], cnt, first])
            if key == "뷰성형외과":
                for c in range(1, 5):
                    ws.cell(row=ws.max_row, column=c).fill = PatternFill("solid", fgColor=WARN)
    for col, w in zip("ABCD", [16, 30, 13, 13]):
        ws.column_dimensions[col].width = w

    # ── 엔진별 상세 (답변 전문 + 인용) ────────────────────
    for sheet, data in (("GPT상세", oa), ("PPLX상세", px)):
        ws = wb.create_sheet(sheet)
        ws.append(["ID", "질문", "뷰 언급", "순위", "답변 전문", "인용 URL"])
        style_header(ws, 6)
        for q in qs:
            r = data[q["id"]]
            ws.append([q["id"], q["question"],
                       "O" if r.get("view_mentioned") else "X", r.get("view_rank") or "",
                       r.get("answer", "")[:32000], "\n".join(r.get("citations", [])[:10])])
            row = ws.max_row
            for c in range(1, 7):
                ws.cell(row=row, column=c).alignment = Alignment(vertical="top", wrap_text=c in (2, 5, 6))
        for col, w in zip("ABCDEF", [8, 36, 8, 6, 90, 50]):
            ws.column_dimensions[col].width = w

    wb.save(OUT)
    print("저장:", OUT)


if __name__ == "__main__":
    main()