MLGE-AC-UFD: Multi-level Graph Embedding and Approximate Computation for Unsupervised Fraud Detection

Yixin Tian, Bingkun Wang, Jiahui Wang, Yilin Huang, Fangshu Chen

Abstract: In the era of big data, online platforms face increasingly sophisticated and covert fraudulent activities that traditional detection methods fail to handle effectively. Traditional methods which depend on predefined rules or supervised learning algorithms, have limitations in capturing the diverse and evolving nature of fraud activity. Additionally, the exponential increase of online users and activities significantly expends the need of computational resources, and hinders promptly detection. More intelligent and efficient detection methods are expected to address these challenges. This paper proposes an unsupervised intelligent fraud detection framework that integrates multi-level graph embedding and approximate computation. By constructing a user behavior graph with both explicit and implicit behaviors, the framework captures comprehensive user activities. Experiment results show that our framework can detect majority of fraudulent behaviors and consequently help companies reduce economic losses. Demo video click here or try onsite.

External IDs:dblp:conf/wise/TianWWHC24