Investigation of Explainable Recommendation based on Virtual User Profiles

Abstract

This paper proposes a method to present virtual user profiles constructed from anonymous data as explanations of the recommendation. Personalized recommender systems have been developed to support users' decision-making. Personalization enables the system to recommend items that the user potentially prefers, but if the user does not accept the recommendation, the user will not be satisfied with the recommendation. In addition, users may feel anxious about their privacy from the highly personalized system, that is, it is supposed that there is a limitation in improving the acceptability depending solely on personalizing the system. In response to these problems, this paper expects that presenting virtual user profiles as an explanation encourages users to accept the recommendation. Based on the analysis of anonymous data, virtual user profiles are constructed, and its validity is evaluated. The evaluation is conducted by a questionnaire in which participants are asked to check the sentences that represent user profiles.