Enhancing the Diversity Adjusting Strategy with Personality Information in Music Recommender Systems

Abstract

Current research on personality and diversity based Recommender Systems (RecSys) are mostly separated. In most diversity-based Recommender Systems, researchers usually endeavored to achieve an optimal balance between accuracy and diversity while they commonly set a same diversity level for all users. Different diversity needs for users with different personalities are rarely studied. Another branch of research on personality-based Recommender Systems mostly emphasize utilizing personality information to enhancing the rating prediction accuracy so as to solve the ’Cold-Start Problem’. While few of them have in depth investigated whether and how it influences users’ other preference needs (such as diversity needs).

This thesis presents the work how we combine these two branches of research together. Anchored in the music domain, we investigate how personality information can be incorporated into the Music Recommender Systems to help adjust the diversity degrees for people with different personalities. We first conducted a pilot study to investigate the correlation between users’ personality factors and their diversity needs on the music recommendations. Results showed that there exits significant correlations between them, especially when we consider the personality factor ’Emotional Stability’. Based on such findings, we then proposed a personality-based diversification algorithm to help enhance the diversity adjusting strategy according to people’s personality information in music recommendations. Our offline and online evaluation results demonstrated that our proposed method is an effective solution to generate personalized recommendation lists with relatively higher diversity.