📄 中文摘要
最近对我们的Telegram反垃圾邮件机器人ModerAI进行了三项重大更改，根本改变了我们处理边缘案例的方式。传统的反垃圾邮件机器人通常做出二元决策：要么判定为垃圾邮件，要么判定为非垃圾邮件。这种方法导致了两个失败模式：一是误判，导致合法用户因在个人简介中包含链接而被封禁；二是漏判，垃圾邮件发送者能够学习规则并规避检测。为了解决这些问题，实施了上下文生物链接分析，通过分析用户简介中的链接上下文，判断其类别，从而更准确地决定是否封禁用户。
构建更公平的反垃圾邮件系统：我们如何处理链接、警告和新聊天
📄 English Summary
Building a Fairer Anti-Spam System: How We Handle Links, Warnings, and New Chats
Three significant changes were recently made to our Telegram anti-spam bot, ModerAI, fundamentally altering how edge cases are handled. Most anti-spam bots operate on binary decisions: spam or not spam. This creates two failure modes: false positives, where legitimate users are banned for having links in their bios, and false negatives, where spammers learn the rules and circumvent them. To address these issues, a contextual bio link analysis was implemented. By analyzing the context of links in user bios, the system can more accurately determine whether to ban a user based on the category of the link.