Defending Multiple-User-Multiple-Target Attacks in Online Reputation Systems

As online reputation systems are playing increasingly important roles in reducing risks of online interactions, attacks against such systems have evolved rapidly. Nowadays, some powerful attacks are conducted by companies that make profit through manipulating reputation of online items for their customers. These items can be products (e.g. in Amazon), businesses (e.g. hotels in travel sites), and digital content (e.g. videos in You tube). In such attacks, colluded malicious users play well-planned strategies to manipulate reputation of multiple target items. To address these attacks, we propose a defense scheme that (1) sets up heterogeneous thresholds for detecting suspicious items and (2) identifies target items based on correlation analysis among suspicious items. The proposed scheme and two other comparison schemes are evaluated by a combination of real user data and simulation data. The proposed scheme demonstrates significant advantages in detecting malicious users, recovering reputation scores of target items, and reducing interference to normal items.