feat: 垃圾信息分类标签功能

新增垃圾信息细分类标签，在朴素贝叶斯二分类基础上对spam进行细分：
- 新增 spam_categorizer.py 分类模块（诈骗/骚扰/广告）
- SpamPredictionLog 和 ContentPost 模型添加 category 字段
- content_routes 和 spam_routes 接口返回分类标签

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

backend/app/routes/content_routes.py

@@ -5,6 +5,7 @@ from flask_jwt_extended import jwt_required
 
 from app.extensions import db
 from app.ml.naive_bayes_classifier import NaiveBayesSpamClassifier
+from app.ml.spam_categorizer import categorize_spam, get_category_label
 from app.models import ContentPost, DetectionConfig, SpamPredictionLog, SpamTrainingSample, User
 from app.utils.auth import current_user
 from app.utils.response import fail, ok
@@ -77,7 +78,7 @@ def _resolve_recipient(payload: dict, visibility: str, current_user_id: int):
     return recipient, None
 
 
-def _predict_and_decide(text: str, user_credit: int = 100) -> tuple[dict, float, bool]:
+def _predict_and_decide(text: str, user_credit: int = 100) -> tuple[dict, float, bool, str, str]:
     """根据用户信誉分调整阈值系数。信誉分越高，阈值越高（降低敏感度）"""
     clf = _ensure_ready()
     result = clf.predict(text)
@@ -92,7 +93,14 @@ def _predict_and_decide(text: str, user_credit: int = 100) -> tuple[dict, float,
 
     adjusted_threshold = base_threshold * credit_factor
     blocked = float(result["spam_probability"]) >= adjusted_threshold
-    return result, adjusted_threshold, blocked
+
+    # 分类标签
+    category = ""
+    category_label = ""
+    if blocked:
+        category, category_label = categorize_spam(result["text"])
+
+    return result, adjusted_threshold, blocked, category, category_label
 
 
 @content_bp.post("/publish")
@@ -113,7 +121,7 @@ def publish_text():
     if err:
         return fail(err, 400)
 
-    result, threshold, blocked = _predict_and_decide(text, user.credit_score or 100)
+    result, threshold, blocked, category, category_label = _predict_and_decide(text, user.credit_score or 100)
 
     post = ContentPost(
         user_id=user.id,
@@ -122,6 +130,7 @@ def publish_text():
         visibility=visibility,
         status="blocked" if blocked else "published",
         prediction=result["prediction"],
+        category=category,
         spam_probability=result["spam_probability"],
         ham_probability=result["ham_probability"],
         confidence=result["confidence"],
@@ -135,6 +144,7 @@ def publish_text():
         user_id=user.id,
         text=result["text"],
         prediction=result["prediction"],
+        category=category,
         spam_probability=result["spam_probability"],
         ham_probability=result["ham_probability"],
         confidence=result["confidence"],
@@ -153,14 +163,18 @@ def publish_text():
 
     db.session.commit()
 
-    feedback = "发布成功" if not blocked else "疑似垃圾信息，系统已拦截，可提交申诉"
+    feedback = "发布成功" if not blocked else f"{category_label or '疑似垃圾信息'}，系统已拦截，可提交申诉"
     return ok(
         {
             "publish_allowed": not blocked,
             "action": "published" if not blocked else "blocked",
             "feedback": feedback,
             "post": _serialize_post(post),
-            "detect": result,
+            "detect": {
+                **result,
+                "category": category,
+                "category_label": category_label,
+            },
         },
         feedback,
     )
@@ -188,13 +202,14 @@ def edit_post(post_id: int):
     if err:
         return fail(err, 400)
 
-    result, threshold, blocked = _predict_and_decide(text, user.credit_score or 100)
+    result, threshold, blocked, category, category_label = _predict_and_decide(text, user.credit_score or 100)
 
     post.text = result["text"]
     post.visibility = visibility
     post.recipient_user_id = recipient.id if recipient else None
     post.status = "blocked" if blocked else "published"
     post.prediction = result["prediction"]
+    post.category = category
     post.spam_probability = result["spam_probability"]
     post.ham_probability = result["ham_probability"]
     post.confidence = result["confidence"]

backend/app/routes/spam_routes.py

@@ -3,6 +3,7 @@ from flask_jwt_extended import jwt_required
 
 from app.extensions import db
 from app.ml.naive_bayes_classifier import NaiveBayesSpamClassifier
+from app.ml.spam_categorizer import categorize_spam, get_category_label
 from app.models import DetectionConfig, SpamPredictionLog, SpamTrainingSample
 from app.utils.auth import admin_required, current_user
 from app.utils.response import fail, ok
@@ -58,10 +59,17 @@ def predict_one():
     threshold = _adjusted_threshold(user.credit_score or 100)
     blocked = float(result["spam_probability"]) >= threshold
 
+    # 分类标签：仅在判定为垃圾时进行细分
+    category = ""
+    category_label = ""
+    if blocked:
+        category, category_label = categorize_spam(result["text"])
+
     row = SpamPredictionLog(
         user_id=user.id,
         text=result["text"],
         prediction=result["prediction"],
+        category=category,
         spam_probability=result["spam_probability"],
         ham_probability=result["ham_probability"],
         confidence=result["confidence"],
@@ -71,7 +79,14 @@ def predict_one():
     db.session.add(row)
     db.session.commit()
 
-    return ok({**result, "log_id": row.id, "threshold": threshold, "blocked_by_threshold": blocked}, "识别成功")
+    return ok({
+        **result,
+        "log_id": row.id,
+        "threshold": threshold,
+        "blocked_by_threshold": blocked,
+        "category": category,
+        "category_label": category_label,
+    }, "识别成功")
 
 
 @spam_bp.post("/predict/batch")
@@ -98,12 +113,23 @@ def predict_batch():
         if len(content) < 2:
             continue
         result = clf.predict(content)
-        result["blocked_by_threshold"] = float(result["spam_probability"]) >= threshold
+        blocked = float(result["spam_probability"]) >= threshold
+        result["blocked_by_threshold"] = blocked
+
+        # 分类标签
+        category = ""
+        category_label = ""
+        if blocked:
+            category, category_label = categorize_spam(result["text"])
+        result["category"] = category
+        result["category_label"] = category_label
+
         rows.append(
             SpamPredictionLog(
                 user_id=user.id,
                 text=result["text"],
                 prediction=result["prediction"],
+                category=category,
                 spam_probability=result["spam_probability"],
                 ham_probability=result["ham_probability"],
                 confidence=result["confidence"],