Enhancing Music Recommendation with User Mood Input

TL;DR

This paper proposes a mood-assisted music recommendation system that uses user mood inputs to improve personalization, demonstrating statistically significant enhancements over baseline methods in user satisfaction.

Contribution

It introduces a novel recommendation approach that incorporates user mood via the energy-valence spectrum, improving music personalization beyond traditional content-based filtering.

Findings

Mood-based recommendations outperform baseline in user ratings

Significant improvement in recommendation relevance

User satisfaction increases with mood integration

Abstract

Recommendation systems have become essential in modern music streaming platforms, due to the vast amount of content available. A common approach in recommendation systems is collaborative filtering, which suggests content to users based on the preferences of others with similar patterns. However, this method performs poorly in domains where interactions are sparse, such as music. Content-based filtering is an alternative approach that examines the qualities of the items themselves. Prior work has explored a range of content-filtering techniques for music, including genre classification, instrument detection, and lyrics analysis. In the literature review component of this work, we examine these methods in detail. Music emotion recognition is a type of content-based filtering that is less explored but has significant potential. Since a user's emotional state influences their musical choices, incorporating user mood into recommendation systems is an alternative way to personalize the listening experience. In this study, we explore a mood-assisted recommendation system that suggests songs based on the desired mood using the energy-valence spectrum. Single-blind experiments are conducted, in which participants are presented with two recommendations (one generated from a mood-assisted recommendation system and one from a baseline system) and are asked to rate them. Results show that integrating user mood leads to a statistically significant improvement in recommendation quality, highlighting the potential of such approaches.