Account Takeover Detection on E-Commerce Platforms

Account takeover is a type of malicious attack where a fraudster steals accounts and passwords from normal users, causing the loss of money and the exposure of personal information. Existing solutions either rely on extensive manual labeling, or require behavior sequences and context graphs of accounts. In this paper, I propose a Siamese neural Network-based Multi-Relation Graph Embedding method (MRGE-SiameseNet) to detect stolen accounts. The key idea of MRGE-SiameseNet is that the two inputs from the same users are similar. I adopt the idea of the siamese neural network to judge whether two different inputs are from the same user. To get a powerful representation of each account, I integrate several embeddings of multiple relationships of accounts and profile feature embedding for each account with multi-head attention mechanism. The fully connected module is employed to obtain the similarity score, which can be utilized to identify whether the account is stolen by account takeover.