Academy of Mathematics and Systems Science, CAS
Colloquia & Seminars

Speaker:	彭浩博士，北京航空航天大学
Inviter:
Title:	Enhancing Graph Neural Network-based Fraud Detectors against Camouflaged Fraudsters
Time & Venue:	2020.08.11 10:00-11:00 腾讯会议 ID：663 213 161
Abstract:	Recently, Graph Neural Networks (GNNs) have been widely applied to fraud detection problems, revealing the suspiciousness of nodes by aggregating their neighborhood information via different relations. However, few prior works have studied the camouflage behavior of fraudsters, which hampers the performance of GNN based fraud detectors during the aggregation process. We introduce two types of camouflages based on recent empirical studies, i.e., the feature camouflage and the relation camouflage. Existing GNNs have not addressed these two camouflages, which results in poor performance of current models. Alternatively, we propose a new model, named as CAmouflage-REsistant GNN (CARE-GNN), to enhance the GNN aggregation process with three modules against camouflages. Concretely, we first devise a label-aware similarity measure to select similar neighbor nodes for aggregation. Then, we leverage reinforcement learning (RL) to adaptively optimize the neighbor filtering threshold for each relation. The optimized thresholds are further combined with the aggregated information w.r.t. relations. Comprehensive experiments on the real-world fraud dataset demonstrate the effectiveness of the RL algorithm. The proposed CARE-GNN also outperforms state-of-the-art GNNs and GNN-based fraud detectors. 报告人简介：彭浩，北京航空航天大学网络空间安全学院讲师，硕士生导师。2019年获北京航空航天大学计算机软件与理论博士学位，曾在美国伊利诺伊大学芝加哥分校访问学习（合作导师Philip S. Yu教授）。研究方向包括社交网络数据表示学习与事件挖掘、时空数据挖掘等，相关研究成果发表在国际顶级学术会议（WWW、SIGIR、AAAI、IJCAI、CIKM等）和知名期刊（IEEE TKDE、IEEE TPDS、IEEE TITS、ACM TSAS、Information Sciences、China Science Information Science、Neurocomputing、KBS、FGCS等），担任人工智能和数据挖掘领域国际会议KDD、IJCAI、AAAI、ACL、CIKM、EMNLP、DASFAA等程序委员；获部级科技奖励一等奖1项（2018年中国电子学会技术发明一等奖，排名第6）；已发表论文38篇，授权国家发明专利11项。