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>
This commit is contained in:
刘正航
2026-04-22 21:52:08 +08:00
parent 84f0943578
commit cedfd066c4
5 changed files with 126 additions and 8 deletions

View File

@@ -0,0 +1,68 @@
"""垃圾信息分类标签模块
在朴素贝叶斯二分类spam/ham基础上对判定为 spam 的文本进行细分类标签。
分类优先级:诈骗 > 骚扰 > 广告(按危害程度排序)
"""
CATEGORY_KEYWORDS = {
"fraud": [
"中奖", "幸运粉丝", "幸运用户", "银行卡异常", "社保异常", "账号冻结",
"解封", "立即验证", "验证码", "欠费停机", "退款待确认", "违章信息",
"紧急通知", "账户异常", "风险", "核验", "被冻结", "将被冻结",
],
"harassment": [
"兼职", "日结", "高薪", "刷单", "赚钱", "外快", "宝妈", "学生都能做",
"添加微信", "扫码进群", "进群立刻", "想赚", "零花钱", "在家办公",
"无需面试", "火热招募", "秒赚", "招募",
],
"advertisement": [
"领取", "优惠", "红包", "优惠券", "秒杀", "返现", "补贴", "会员",
"特价", "低价", "点击链接", "扫码", "免费领取", "无门槛", "现金券",
"盲盒", "百分百中奖", "隐藏优惠券", "内部价", "货到付款", "限时",
"最后", "名额", "先到先得",
],
}
CATEGORY_LABELS = {
"fraud": "疑似诈骗",
"harassment": "疑似骚扰",
"advertisement": "疑似广告",
"spam": "疑似垃圾",
"ham": "",
}
CATEGORY_PRIORITY = ["fraud", "harassment", "advertisement"]
def categorize_spam(text: str) -> tuple[str, str]:
"""根据关键词匹配判定垃圾信息的具体分类标签
Args:
text: 待分类的文本内容
Returns:
tuple[str, str]: (category_code, category_label)
- category_code: fraud | harassment | advertisement | spam
- category_label: 疑似诈骗 | 疑似骚扰 | 疑似广告 | 疑似垃圾
"""
text_lower = text.lower()
for category in CATEGORY_PRIORITY:
keywords = CATEGORY_KEYWORDS.get(category, [])
for kw in keywords:
if kw.lower() in text_lower:
return category, CATEGORY_LABELS[category]
return "spam", CATEGORY_LABELS["spam"]
def get_category_label(category: str) -> str:
"""获取分类标签的中文显示文本
Args:
category: 分类代码
Returns:
str: 中文标签文本
"""
return CATEGORY_LABELS.get(category, "")

View File

@@ -79,6 +79,7 @@ class SpamPredictionLog(db.Model):
user_id = db.Column(db.Integer, db.ForeignKey("users.id"), nullable=False, index=True)
text = db.Column(db.Text, nullable=False)
prediction = db.Column(db.String(16), nullable=False) # spam | ham
category = db.Column(db.String(32), default="") # fraud | harassment | advertisement | spam | 空
spam_probability = db.Column(db.Float, nullable=False)
ham_probability = db.Column(db.Float, nullable=False)
confidence = db.Column(db.Float, nullable=False)
@@ -92,6 +93,7 @@ class SpamPredictionLog(db.Model):
"user_id": self.user_id,
"text": self.text,
"prediction": self.prediction,
"category": self.category or "",
"spam_probability": round(float(self.spam_probability), 4),
"ham_probability": round(float(self.ham_probability), 4),
"confidence": round(float(self.confidence), 4),
@@ -130,6 +132,7 @@ class ContentPost(db.Model):
status = db.Column(db.String(16), nullable=False, default="published") # published | blocked
prediction = db.Column(db.String(16), nullable=False, default="ham")
category = db.Column(db.String(32), default="") # fraud | harassment | advertisement | spam | 空
spam_probability = db.Column(db.Float, nullable=False, default=0)
ham_probability = db.Column(db.Float, nullable=False, default=0)
confidence = db.Column(db.Float, nullable=False, default=0)
@@ -163,6 +166,7 @@ class ContentPost(db.Model):
"visibility": self.visibility,
"status": self.status,
"prediction": self.prediction,
"category": self.category or "",
"spam_probability": round(float(self.spam_probability), 4),
"ham_probability": round(float(self.ham_probability), 4),
"confidence": round(float(self.confidence), 4),

View File

@@ -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"]

View File

@@ -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"],

View File

@@ -0,0 +1,5 @@
-- 添加 category 字段到 spam_prediction_logs 表
ALTER TABLE `spam_prediction_logs` ADD COLUMN `category` VARCHAR(32) DEFAULT '' AFTER `prediction`;
-- 添加 category 字段到 content_posts 表
ALTER TABLE `content_posts` ADD COLUMN `category` VARCHAR(32) DEFAULT '' AFTER `prediction`;