Detecting unfair rating attacks in online rating systems

Abstract

Ecommerce sites, such as EBay and Amazon, adopt rating systems that assist buyers in selecting reliable sellers to transact with. Due to intense competition among sellers, rating systems may be faced with unfair rating attacks, which can boost or tarnish a seller’s reputation. Rating systems thus need to be robust against attacks, so that reliable information is provided to consumers. A system has been developed, to simulate the behavior of buyers, sellers, and transactions that occur in e-­‐marketplaces, where ratings may be based on a single-­‐ criterion or multi-­‐criterions. In this report, 4 experiments are conducted, (1) single-­‐criteria, simulated environment, (2) single-­‐criteria, real environment with simulated attacks, (3) multi-­‐criteria, simulated environment, and (4) multi-­‐criteria, real environment with simulated attacks. These experiments are performed using 4 defense models (BRS, iCLUB, EBay, MeTrust) against various unfair rating attacks. An analysis is done on the performance of each model, which is benchmarked using 3 evaluation metrics.