Proposal on Virtual User Profile Generation for Explainable Recommendation

Proposal on Virtual User Profile Generation for Explainable Recommendation

연구 분야: Cryptography

학회: International Symposium on Computational Intelligence and Industrial Applications

This paper proposes a method for providing users with the profiles of virtual users as an explanation for recommendations. Recommender systems are one of the intelligent systems that support us in accessing vast amounts of information. Those are roughly divided into content-based filtering and collaborative filtering. Many algorithms have been studied for determining items to recommend, all of which require users’ personal information such as purchase/browsing history to estimate their tastes for providing personalized recommendations. Although obtaining as much information as possible is preferable, it raises the problem of privacy concerns. To realize personalized recommendations without collecting users’ personal information, this paper proposes a recommendation framework that uses virtual user profiles. A virtual user profile describes the interests and tastes of a virtual user to items. A virtual user is extracted from large interaction data about anonymous users. Using the profiles of virtual users and their ratings to items of interest as a kind of explanation for recommendations, users are expected to find relevant items without providing their private information. This paper describes how to create a virtual user profile and shows its effectiveness through questionnaires.