Enhancing Long-Tail Artist Recommendation with a Relevance, Diversity, and Unexpectedness Channel Framework

DOI

Bibliographic Information

Other Title
  • 関連性・多様性・意外性チャネルフレームワークによるロングテールアーティスト推薦の強化

Search this article

Description

Serendipitous recommendations have attracted more attention since considering solely accuracy may be insufficient to satisfy users' needs. According to data from Rolling Stone, 90% of music listening on Spotify is attributed to only the top 1% of all artists, highlighting a popularity bias issue that significantly impacts recommendations for long-tailed artists. Furthermore, relying solely on a user's past listening history may lead to the filtering bubble effect, limiting the discovery of diverse and unexpected artists that the user may prefer. In this work, we propose a novel multi-channel artist self-attention model for playlist-artist recommendations that considers three aspects of serendipity: relevance, diversity, and unexpectedness. The relevance channel aims to suggest similar artists that frequently appear together in playlists, the diversity channel aims to explore more diverse artists in terms of genres, and the unexpectedness channel aims to reveal artists with similar detailed descriptions but are least expected. Our extensive experiments, compared with state-of-the-art baselines, demonstrate that incorporating serendipity channels not only enhances the accuracy of recommendations but also promotes the discovery of suitable long-tailed artists.

Journal

Keywords

Details 詳細情報について

Report a problem

Back to top