Pierre Hanna

Pierre Hanna

The consumption of music has its specificities in comparison with other media, especially in relation to listening durations and replays. Music recommendation can take these properties into account in order to predict the behaviours of the users. Their impact is investigated in this paper. A large database was thus created using logs collected on a streaming platform, notably collecting the listening times. The proposed study shows that a high proportion of the listening events implies a skip action, which may indicate that the user did not appreciate the track listened. Implicit like and dislike can be deduced from this information of durations and replays and can be taken into account for music recommendation and for the evaluation of music recommendation engines. A quantitative study as usually found in the literature confirms that neighborhood-based systems considering binary data give the best results in terms of MAP@k. However, a more qualitative evaluation of the recommended tracks shows that many tracks recommended, usually evaluated in a positive way, lead to skips or thus are actually not appreciated. We propose the consideration of implicit like/dislike as recommendation engine inputs. Evaluations show that neighbourhood-based engines remain the most precise, but filtering inputs according to durations and/or replays have a significant positive impact on the objective of the recommendation engine. The recommendation process can thus be improved by taking account of listening durations and replays. We also study the possibility of post-filtering a list of recommended tracks so as to limit the number of tracks that will be unpleasantly listened (skip and implicit dislike) and to increase the proportion of tracks appreciated (implicit like). Several simple algorithms show that this post-filtering operation leads to an improvement of the quality of the music recommendations.

Yann Bayle, Matthias Robine, Pierre Hanna

This study deals with content-based musical playlists generation focused on Songs and Instrumentals. Automatic playlist generation relies on collaborative filtering and autotagging algorithms. Autotagging can solve the cold start issue and popularity bias that are critical in music recommender systems. However, autotagging remains to be improved and cannot generate satisfying music playlists. In this paper, we suggest improvements toward better autotagging-generated playlists compared to state-of-the-art. To assess our method, we focus on the Song and Instrumental tags. Song and Instrumental are two objective and opposite tags that are under-studied compared to genres or moods, which are subjective and multi-modal tags. In this paper, we consider an industrial real-world musical database that is unevenly distributed between Songs and Instrumentals and bigger than databases used in previous studies. We set up three incremental experiments to enhance automatic playlist generation. Our suggested approach generates an Instrumental playlist with up to three times less false positives than cutting edge methods. Moreover, we provide a design of experiment framework to foster research on Songs and Instrumentals. We give insight on how to improve further the quality of generated playlists and to extend our methods to other musical tags. Furthermore, we provide the source code to guarantee reproducible research.