P2P Meta-Recommenders: Aggregated Diversity Maximization as a Bulwark against Attacks on Reviewers

摘要

We focus on the problem of selecting reviewers (or raters) that are considered by a recommender system (or a user) under the aspect of security. Malicious reviewers can exert unreasonable influence, and can bias online consumers unfairly against an attacked item or competitor. This paper proposes an approach where a meta-recommender maximizes the aggregated diversity of reviewers when deciding which reviews should be considered by a recommender system or an online consumer. This problem can be of interest in many domains where producers or service providers may seek advantages by compromising competitors with fake reviews or ratings such as tourism and hospitality industries or even free open-source software. A solution is proposed for users linked in social networks, such as unstructured P2P societies. In order to evaluate the proposed solution, we describe a mechanism of selecting reviewers of software updates such that not all end-users of a software are impacted by a potentially malicious strict subset of all available reviewers, and we experimentally assess resistance to attacks.