Update filter

[yzqzss]

把 xlog 全部4万多文章拉了下来，打标然后训练了个简单的 TF-IDF spam 分类器。
在最新的3000篇（跨度3个月）的博客文章上跑了下预测。

以下是这3000篇中新发现的 47 个至少有一篇被预测为 spam 文章的账号的情况：

未被勾选的为实实在在的 spam 账号（已人工复检）。
勾选的为识别错误的账号（已人工复检）。

• 识别为spam文章数量,识别为非spam文章数量,用户ID
• 101,0,https://xlog.app/api/redirection?characterId=76350
• 93,0,https://xlog.app/api/redirection?characterId=75664
• 71,3,https://xlog.app/api/redirection?characterId=75820
• 65,3,https://xlog.app/api/redirection?characterId=76863
• 50,0,https://xlog.app/api/redirection?characterId=75947
• 28,0,https://xlog.app/api/redirection?characterId=75543
• 26,0,https://xlog.app/api/redirection?characterId=76286
• 24,0,https://xlog.app/api/redirection?characterId=76192
• 24,0,https://xlog.app/api/redirection?characterId=75803
• 23,0,https://xlog.app/api/redirection?characterId=75659
• 17,0,https://xlog.app/api/redirection?characterId=75414
• 14,0,https://xlog.app/api/redirection?characterId=76892
• 13,0,https://xlog.app/api/redirection?characterId=76060
• 11,0,https://xlog.app/api/redirection?characterId=76196
• 11,0,https://xlog.app/api/redirection?characterId=74917
• 10,0,https://xlog.app/api/redirection?characterId=76893
• 10,0,https://xlog.app/api/redirection?characterId=76197
• 8,1,https://xlog.app/api/redirection?characterId=76421
• 6,28,https://xlog.app/api/redirection?characterId=77278 # sex spam
• 6,0,https://xlog.app/api/redirection?characterId=76993
• 6,0,https://xlog.app/api/redirection?characterId=76991
• 6,0,https://xlog.app/api/redirection?characterId=76989
• 5,0,https://xlog.app/api/redirection?characterId=77150
• 4,1,https://xlog.app/api/redirection?characterId=76652
• 4,0,https://xlog.app/api/redirection?characterId=76674
• 3,4,https://xlog.app/api/redirection?characterId=76411 # sex spam
• 3,3,https://xlog.app/api/redirection?characterId=74718 # 2025-05-13 发了 6 篇无意义的文章，又全删了，也许是测试？应该是真人。
• 3,19,https://xlog.app/api/redirection?characterId=76729 # sex spam
• 3,12,https://xlog.app/api/redirection?characterId=75979 # 二次元涩涩，好👌
• 3,0,https://xlog.app/api/redirection?characterId=77178
• 3,0,https://xlog.app/api/redirection?characterId=77035
• 3,0,https://xlog.app/api/redirection?characterId=76402
• 2,0,https://xlog.app/api/redirection?characterId=76735
• 2,0,https://xlog.app/api/redirection?characterId=76551
• 1,8,https://xlog.app/api/redirection?characterId=68402
• 1,70,https://xlog.app/api/redirection?characterId=75963 # 虽然有 70 篇文章未被检出 spam,但人工复查发现所有的文章都是“xxxx项目数据怎么看？”这种形式的 AI 生成文章。
• 1,7,https://xlog.app/api/redirection?characterId=77277 # 真人用户，唯一被判定为 spam 的文章是因为脏话占比有点高，所以被判定为 sex spam 了，置信度只有 0.59 。
• 1,3,https://xlog.app/api/redirection?characterId=77121 # 真人用户
• 1,2,https://xlog.app/api/redirection?characterId=76744 # 真人用户
• 1,1,https://xlog.app/api/redirection?characterId=76299
• 1,1,https://xlog.app/api/redirection?characterId=72236
• 1,0,https://xlog.app/api/redirection?characterId=77156
• 1,0,https://xlog.app/api/redirection?characterId=76740
• 1,0,https://xlog.app/api/redirection?characterId=76604 # sex spam
• 1,0,https://xlog.app/api/redirection?characterId=76287
• 1,0,https://xlog.app/api/redirection?characterId=76049
• 1,0,https://xlog.app/api/redirection?characterId=73768 # 这个用户好像在把 xlog 当图床用

这样就剩下了 41 个确定的近三个月新出现的 spam 账号，加进 filter 里了。

---

以及3000篇文章每篇的详细结果:

3000_prediction_results.csv

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

1 Skipped Deployment

Name	Status	Preview	Comments	Updated (UTC)
xlog	⬜️ Ignored (Inspect)	Visit Preview		Jun 15, 2025 10:38am