时间：2026年1月13日 下午14：45
地点：杨咏曼606会议室
报告内容简介：Phishing remains a pervasive social-engineering threat, causing substantial financial and security harm. Its practical detection is challenging: phishing webpages are easy to create, attackers rapidly adapt to evade defenses, and sites are often short-lived—requiring anti-phishing systems to be both cost-effective and highly accurate in real time.
Many existing approaches struggle in deployment. They frequently rely on manually engineered features that can be brittle under distribution shift, provide limited explainability for analyst trust and actionability, and operate reactively rather than discovering new threats early.
We present a proactive phishing defense framework that combines an explainable, reference-based detector with an end-to-end monitoring system. Our detector identifies the impersonated brand via comparison to a protected reference set, localizes and categorizes credential-collection regions, and uses counterfactual web testing to expose suspicious behaviors. Notably, it generalizes to real-world phishing without training on collected phishing webpages, while producing interpretable evidence to support verification and response. Building on this detector, our monitoring pipeline performs large-scale crawling, incident reporting, and trend visualization, generating on average 50 zero-day phishing alerts per day and providing actionable insights into evolving phishing tactics.
报告人简介：刘若凡，新加坡国立大学计算机博士后研究员，研究方向包括人工智能在网络安全与Web欺诈检测中的应用。2021–2024年在新加坡国立大学攻读博士学位，师从Dong Jin Song教授，期间开展多个关于可解释钓鱼检测的研究。曾获得NUS Dean’s Graduate Research Excellence Award、NUS Research Achievement Award等多项学术荣誉，中国国际大学生创新大赛2025（互联网+）国际项目东南亚赛区第一名，并在USENIX Security、ISSTA等CCF-A类国际会议发表11篇论文，贡献包括基于参考的钓鱼检测、GUI测试等。其研发的多项工具（如Phishpedia、PhishIntention、DynaPhish、PhishLLM）在钓鱼监测与安全分析领域具有实践影响。
