Semantics-Aware Shilling Attacks against collaborative recommender systems via Knowledge Graphs

Tracking #: 2735-3949

This paper is currently under review
Vito Walter Anelli
Yashar Deldjoo
Tommaso Di Noia
Eugenio Di Sciascio
Felice Antonio Merra

Responsible editor: 
Guest Editors ESWC 2020

Submission type: 
Full Paper
Several domains have widely benefited from the adoption of Knowledge graphs (KGs). For recommender systems (RSs), the adoption of KGs has resulted in accurate, personalized recommendations of items/products according to users' preferences. Among different recommendation techniques, collaborative filtering (CF) is one the most promising approaches to build RSs. Their success is due to the effective exploitation of similarities/correlations encoded in user interaction patterns. Nonetheless, their strength is also their weakness. A malicious agent can add fake user profiles into the platform, altering the genuine similarity values and the corresponding recommendation lists. While the research community has extensively studied KGs to solve various recommendation problems, sufficient attention was not paid to the possibility of exploiting KGs to compromise the quality of recommendations. KGs provide a rich source of information for item representation and recommendation that can dramatically increase the attackers' knowledge about the victim recommendation platform. To this end, this article introduces a new attack strategy, named semantics-aware shilling attack (SAShA), that leverages semantic features extracted from a knowledge graph. SAShA provides the semantics-aware variant of three state-of-the-art attack strategies: Random, Average, and BandWagon. These improved attacks can exploit graph relatedness measures, i.e., Katz and Exclusivity-based, computed considering 1-hop and 2-hops of graph exploration. We performed an extensive experimental evaluation with four state-of-the-art recommendation systems and two well-known recommendation datasets to investigate the effectiveness of SAShA. Since the semantics of relations has a crucial role in KGs, we have also analyzed the impact of relations' semantics by grouping them in various classes. Experimental results indicate the benefit of embracing KGs in favor of the attackers' capability in attacking recommendation systems.
Full PDF Version: 
Under Review