Abstract
The rating fraud in online e-commerce stores targets at receiving large revenues through boosting the popularity of selected items with fake ratings. The challenges of detecting rating frauds come from discovering small scale abnormal activities in a large amount of data and detecting frauds in a time-critical manner from online rating streams. This paper presents a real-time visual analytics system that consists of two essential components: a server for automatically handling data streams and a visual analytics interface for performing interactive analysis. Based on the features of rating frauds, we present a detection solution which balances computationally expensive algorithms and interactive analysis between the server and analysts. Specifically, our detection system filters data through performing an initial suspicion level detection on the server, and analysts can combine different statistical analysis of the user / item matrix through a co-mapped singular value decomposition (SVD) diagram, re-ordered matrix representation, and the temporal view. We demonstrate our approach with case studies of different fraud scenarios and show that rating frauds can be effectively detected.