Enhancing Serendipity by Incorporating User Control of Social Information in a Recommender System

초록

Modern recommender systems often prioritize accuracy, resulting in repetitive recommendations that hinder user experience. As an alternative, serendipity has been considered for recommender systems to promote novel and unexpected yet relevant recommendations; however, balancing its core elements under evolving user preferences remains challenging. To address this challenge, we propose a novel interactive recommender system that incorporates user-selected social information to generate serendipitous recommendations. Specifically, input features are categorized as static or dynamic, and the system integrates dynamic features from user-controlled social information into the user profile. This integration is managed through a personalization parameter that acts as a weighted average, determining how strongly the selected user information influences the final recommendations. Scenario-based experiments on datasets show that the proposed system achieves balanced gains in serendipity across user profiles. This study contributes to recommender system research by enabling user control over social-information influence, yielding more contextually aligned and personally meaningful recommendations.

키워드