A novel similarity-based taste features-extracted emotions-aware music recommendation algorithm

Abstract

Human music tastes are subjective and difficult to measure, with existing recommendation algorithms often failing to consider music similarity, taste features, and emotions simultaneously. This paper proposes a novel music recommendation algorithm that integrates music similarity, taste features, and emotions, organized into five modules. Motivated by the probabilistic linguistic term set (PLTS), we establish attribute feature vectors of songs by associating attribute values with corresponding probabilities. Module 1 establishes behavior matrix of users to calculate comprehensive behavioral feeling score for obtaining the original music interests of users. Module 2 designs two improved collaborative filtering algorithms to alleviate data sparsity of intuitionistic fuzzy music taste matrix. Module 3 extracts taste features from user listening behavior and the favorite songs of users. Module 4 integrates subjective and objective music emotions to obtain the comprehensive music emotions of user. Considering the dynamic change in users’ music taste features, we incorporate the latest taste features in Module 5 to reorder the song list obtained by the above four modules. The experiment results verify the effectiveness of this recommendation algorithm. It significantly outperforms three popular music recommendation systems in accuracy, excels in ranking quality, new song accuracy, and richness metrics, marginally surpasses them in listening duration.