From Explanation to Exploration: Promoting DivErsity in Recommendation Systems

Antonino Ferraro, Antonio Galli, Valerio La Gatta, Marco Postiglione, Diego Russo, Gian Marco Orlando, Giuseppe Riccio, Antonio Romano, Vincenzo Moscato

Abstract: Contemporary recommender systems excel in personalizing content based on user preferences, yet concerns persist regarding recommendation diversity and the ensuing rabbit hole effect. Prior research on explanation-driven diversification indicates the importance of analyzing the similarity of explanations for different items within a single recommendation list to promote recommendation diversity. However, it is equally crucial to assess each item’s potential impact on guiding recommendations towards a rabbit hole over time. This paper delves into the dynamic relationship between explanations and diversity across multiple interactions within a recommender system. We introduce FEEDERS (From Explanation to Exploration: promoting DivErsity in Recommendation Systems), a framework designed to leverage explanations for enhancing recommendation diversity and mitigating the rabbit hole effect in repeated interactions with a recommender system. Specifically, FEEDERS leverages an explainable recommender system and enables multiple corrective actions, such as item re-ranking and modification of item importance before recommendation. Our experiments using the PGPR framework [39] and the MovieLens1M dataset demonstrate FEEDERS’ effectiveness in enhancing recommendation diversity, achieving a notable up to 10% increase compared to its closest competitor, while maintaining competitive accuracy performance. In contrast to all baselines, FEEDERS consistently exhibits or increases diversity trends across consecutive interactions with the recommender system, indicating resilience against the rabbit hole effect.

External IDs:doi:10.1007/978-3-031-87654-7_13