On the negative impact of social influence in recommender systems: A study of bribery in collaborative hybrid algorithms

Guilherme Ramos and Carlos Caleiro (University of Lisbon, Portugal) and Ludovico Boratto (EURECAT (Centre Tecnológic de Catalunya), Barcelona, Spain) published in Information Processing and Management Volume 57 Issue 2 (March 2020)

Highlights

• We propose a novel hybrid Collaborative Filtering algorithm to counter bribing.
• We identify, from the point of view of a seller of an item, which users are profitable to bribe.
• We show that our algorithm is as effective as the state-of-the-art approaches, while being more efficient.
• We illustrate our framework, by studying the impact of bribing in our algorithm and a real-world system.

Abstract

Recommender systems are based on inherent forms of social influence. Indeed, suggestions are provided to the users based on the opinions of peers. Given the relevance that ratings have nowadays to push the sales of an item, sellers might decide to bribe users so that they rate or change the ratings given to items, thus increasing the sellers’ reputation.

Hence, by exploiting the fact that influential users can lead an item to get recommended, bribing can become an effective way to negatively exploit social influence and introduce a bias in the recommendations.

Given that bribing is forbidden but still employed by sellers, we propose a novel matrix completion algorithm that performs hybrid memory-based collaborative filtering using an approximation of Kolmogorov complexity. We also propose a framework to study the bribery effect and the bribery resistance of our approach.

Our theoretical analysis, validated through experiments on real-world datasets, shows that our approach is an effective way to counter bribing while, with state-of-the-art algorithms, sellers can bribe a large part of the users.

Labels:
bribing, algorithmic_bias, social_influence.

---