ai-labeling: Prompt 1 v0.1 + ground-truth v0.4.0(20期) + MiMo首轮跑批

- Prompt 1 v0.1: 4分类字段(题材/装备域/场景/技术)判别prompt，含5示例+边界规则+换装测试
- ground-truth v0.4.0: 10→20期扩展，ep编号重映射(旧ep3-15→新ep001-020)，4分类字段全填(制片人逐期审定)
- 文稿升级: 旧10期删除，新20期md文件(doco子项目清洗产出)替换
- 脚本升级: ep编号2位→3位，ALL_EPISODES扩至1-20，summarize classification比对逻辑实现
- MiMo首轮结果: 题材75%/装备域70%/场景95%/技术70%/全对40%，需迭代Prompt
- 质量标杆: 加入PPT+Excel样板文件(example/)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

ai-labeling/scripts/summarize.py

@@ -28,7 +28,7 @@ def latest_per_ep(files):
     latest = {}
     for f in files:
         name = f.name
-        # 文件名格式: 20260611_154037_m3_ep04.json
+        # 文件名格式: 20260611_154037_model_field_ep003.json
         parts = name.replace(".json", "").split("_")
         if len(parts) >= 4:
             ts = parts[0] + parts[1]  # yyyymmddHHMMSS
@@ -79,8 +79,19 @@ def run(model, field="narrative"):
             hit = pred_val == gt_val if pred_val is not None else False
             rows.append({"ep": ep, "title": title, "gt": gt_val, "pred": pred_val or "解析失败", "hit": hit, "conf": conf})
         elif field == "classification":
-            # 骨架预留：等 ground-truth 标注就绪后实现具体比对逻辑
-            rows.append({"ep": ep, "title": title, "gt": "?", "pred": "classification 待实现", "hit": False, "conf": "?"})
+            conf = result.get("confidence", "?") if result else "?"
+            sub = {}
+            for key in ("program_format", "equipment_domain", "scene_tags", "tech_tags"):
+                gt_val = gt.get(key)
+                pred_val = result.get(key) if result else None
+                if gt_val is None or pred_val is None:
+                    sub[key] = None
+                elif isinstance(gt_val, list):
+                    sub[key] = set(gt_val) == set(pred_val) if isinstance(pred_val, list) else False
+                else:
+                    sub[key] = pred_val == gt_val
+            all_hit = all(v is True for v in sub.values() if v is not None)
+            rows.append({"ep": ep, "title": title, "sub": sub, "all_hit": all_hit, "conf": conf})
 
     if field == "narrative":
         # 打印每行
@@ -103,8 +114,25 @@ def run(model, field="narrative"):
         print(f"  自评\"中/低\"置信的命中率: {ml_hit}/{len(mid_low)}")
         print(f"  解析失败: {parse_fail} 期")
     elif field == "classification":
-        print(f"\n  ===== {model} classification =====")
-        print(f"  classification 比对逻辑待实现（等 ground-truth 标注就绪）")
+        field_names = ["program_format", "equipment_domain", "scene_tags", "tech_tags"]
+        short = {"program_format": "题材", "equipment_domain": "装备域", "scene_tags": "场景", "tech_tags": "技术"}
+        for r in rows:
+            marks = ""
+            for fn in field_names:
+                v = r["sub"].get(fn)
+                marks += "✓" if v is True else ("✗" if v is False else "-")
+            all_mark = "✓" if r["all_hit"] else "✗"
+            print(f'  ep{r["ep"]:03d} {r["title"]:<12} | {"/".join(short[f] for f in field_names)}={marks} | 全对:{all_mark} 置信度:{r["conf"]}')
+
+        total = len(rows)
+        for fn in field_names:
+            scored = [r for r in rows if r["sub"].get(fn) is not None]
+            hits = sum(1 for r in scored if r["sub"][fn])
+            pct = hits * 100 // len(scored) if scored else 0
+            print(f"  {short[fn]}命中: {hits}/{len(scored)} = {pct}%")
+        all_scored = [r for r in rows if any(v is not None for v in r["sub"].values())]
+        all_hits = sum(1 for r in all_scored if r["all_hit"])
+        print(f"  4字段全对: {all_hits}/{len(all_scored)} = {all_hits*100//len(all_scored) if all_scored else 0}%")
         print(f"  解析失败: {parse_fail} 期")