C.E.L_Slide_test2/tests/matching/structure_v3.py

"""Phase 21b용 구조 ontology 기반 매칭
- detect_mdx_structure_v3(): MDX 본문에서 같은 schema 속성 emit
- structural_match_v3(): 속성 교집합 기반 score (0~1)
- 가중치는 tie-breaker 수준 (0.05~0.10)
- confidence high 일 때만 점수 부여
"""
import re
import yaml
from pathlib import Path

HERE = Path(__file__).parent


def load_ontology():
    p = HERE / "structure_ontology.yaml"
    with open(p, encoding="utf-8") as f:
        return yaml.safe_load(f)["frames"]


# ═══ MDX 구조 감지 v3 ═══
def detect_mdx_structure_v3(text, mdx_title=""):
    """schema emit: {family, surface, columns, items, has_table, has_cards, semantic_role, confidence}"""
    lines = text.split("\n")
    result = {
        "family": None, "surface": None, "semantic_role": None,
        "columns": None, "items": None,
        "has_table": False, "has_cards": False, "has_diagram": False,
        "confidence": "low",
    }

    # 1. ### X.Y 서브섹션 감지
    subs = [ln for ln in lines if re.match(r"^###\s+\d+\.\d+", ln)]
    subsection_texts = [ln for ln in subs]
    n_subs = len(subs)

    # 2. 표 감지
    table_header = None
    for ln in lines:
        stripped = ln.strip()
        if re.match(r"^\|.*\|.*\|", stripped) and not re.match(r"^\|[\s\-:]+\|", stripped):
            table_header = stripped
            break

    # 3. 최상위 볼드 블릿 수
    top_bullets = [ln for ln in lines if re.match(r"^[-*]\s+\*\*", ln)]
    n_bullets = len(top_bullets)

    full_text = (text + " " + mdx_title).lower()

    # ═══ semantic_role 감지 (figma_audit과 동일 — best-match + priority) ═══
    from figma_audit import _SEMANTIC_HINT_PATTERNS, _ROLE_PRIORITY
    candidates = []
    for role_name, req_kws, min_req in _SEMANTIC_HINT_PATTERNS:
        hits = sum(1 for k in req_kws if k in full_text)
        if hits >= min_req:
            ratio = hits / len(req_kws)
            candidates.append((hits, ratio, role_name))
    if candidates:
        # 1) hits 많음 2) ratio 높음 3) priority 높음
        candidates.sort(key=lambda x: (-x[0], -x[1], -_ROLE_PRIORITY.get(x[2], 0)))
        result["semantic_role"] = candidates[0][2]
    else:
        result["semantic_role"] = None

    # ═══ family / surface / columns 분기 ═══
    # 우선순위: ### 서브섹션 > 표 > 블릿
    if n_subs >= 2:
        # 과정/결과 비교 → compare family
        sub_lower = " ".join(subsection_texts).lower()
        if any(kw in sub_lower for kw in ["과정", "결과", "process", "product"]):
            result["family"] = "compare"
            result["surface"] = "banner-plus-2col"
            result["columns"] = 2
            result["items"] = 2
            result["confidence"] = "high"
        elif n_subs == 2:
            result["family"] = "compare"
            result["surface"] = "2col-paired"
            result["columns"] = 2
            result["items"] = 2
            result["confidence"] = "high"
        else:
            result["family"] = "list"
            result["surface"] = "cards-3-horizontal"
            result["columns"] = n_subs
            result["items"] = n_subs
            result["confidence"] = "medium"
    elif table_header:
        result["has_table"] = True
        cols = [c.strip().replace("*", "").lower() for c in table_header.strip("|").split("|") if c.strip()]
        n_cols = len(cols) - 1  # 첫 열 = label
        col_text = " ".join(cols)
        if any(kw in col_text for kw in ["발주자", "시공자", "설계자"]):
            result["family"] = "cards"
            result["surface"] = "table-persona-3col"
            result["semantic_role"] = result.get("semantic_role") or "persona-benefits"
            result["columns"] = 3
            result["items"] = 3
            result["has_cards"] = True
            result["confidence"] = "high"
        elif any(kw in col_text for kw in ["제조업", "건축", "토목"]) and "토목" in col_text:
            result["family"] = "table"
            result["surface"] = "table-3col"
            result["columns"] = 3
            result["items"] = 3
            result["confidence"] = "high"
        elif ("bim" in col_text) and ("dx" in col_text):
            result["family"] = "compare"
            result["surface"] = "table-multi-row"
            result["columns"] = 2
            result["confidence"] = "high"
        else:
            result["family"] = "table"
            result["surface"] = f"table-{n_cols}col"
            result["columns"] = n_cols
            result["confidence"] = "medium"
    elif n_bullets >= 1:
        result["has_cards"] = True
        if n_bullets == 3:
            result["family"] = "list"
            result["surface"] = "bullets-3"
            result["columns"] = 3
            result["items"] = 3
            result["confidence"] = "high"
        elif n_bullets == 2:
            result["family"] = "compare"
            result["surface"] = "bullets-2col"
            result["columns"] = 2
            result["items"] = 2
            result["confidence"] = "medium"
        elif n_bullets == 4:
            result["family"] = "cards"
            result["surface"] = "cards-4-grid"
            result["columns"] = 4
            result["items"] = 4
            result["confidence"] = "medium"
        else:
            result["family"] = "list"
            result["surface"] = "mixed-bullets"
            result["items"] = n_bullets
            result["confidence"] = "low"
    else:
        result["family"] = None
        result["surface"] = "single-column"
        result["confidence"] = "low"

    return result


# ═══ 구조 매칭 v3 ═══
def structural_match_v3(mdx_schema, fig_schema):
    """속성 교집합 기반 score (0~1). confidence high + 주요 속성 일치시에만 점수 부여.
    Tie-breaker 용도라 가중치는 호출 측에서 낮게 적용 (0.05~0.10 권장)."""
    if not mdx_schema or not fig_schema:
        return 0.0, []
    # confidence 체크: low면 신뢰 안 함
    if mdx_schema.get("confidence") == "low":
        return 0.0, ["mdx confidence low → 0"]
    # 속성 5개 비교
    aligns = []
    scores = []
    # family (가장 중요)
    m_fam = mdx_schema.get("family")
    f_fam = fig_schema.get("family")
    if m_fam and f_fam and m_fam == f_fam:
        scores.append(0.40)
        aligns.append(f"family:{m_fam}✓")
    elif m_fam and f_fam:
        aligns.append(f"family:{m_fam}≠{f_fam}")
    # columns / items
    m_cols = mdx_schema.get("columns") or mdx_schema.get("items")
    f_cols = fig_schema.get("columns") or fig_schema.get("items")
    if m_cols and f_cols and m_cols == f_cols:
        scores.append(0.25)
        aligns.append(f"columns:{m_cols}✓")
    elif m_cols and f_cols:
        aligns.append(f"columns:{m_cols}≠{f_cols}")
    # has_table
    if mdx_schema.get("has_table") is not None and fig_schema.get("has_table") is not None:
        if mdx_schema["has_table"] == fig_schema["has_table"]:
            scores.append(0.10)
            aligns.append(f"has_table:{mdx_schema['has_table']}✓")
    # has_cards
    if mdx_schema.get("has_cards") is not None and fig_schema.get("has_cards") is not None:
        if mdx_schema["has_cards"] == fig_schema["has_cards"]:
            scores.append(0.10)
            aligns.append(f"has_cards:{mdx_schema['has_cards']}✓")
    # semantic_role (bonus)
    m_role = mdx_schema.get("semantic_role")
    f_role = fig_schema.get("semantic_role")
    if m_role and f_role and m_role == f_role and m_role != "(검수 필요)":
        scores.append(0.15)
        aligns.append(f"role:{m_role}✓")
    total = sum(scores)
    return total, aligns