Recommender System using Audio and Lyrics

Abstract

Music streaming services have become an essential part of our daily life. These platforms' recommendation systems are essential because they let consumers receive tailored music recommendations. Similar songs can be found using content-based recommendation systems that make use of audio attributes and lyrics. Major music streaming services, however, mostly rely on audio characteristics. This study proposes a novel approach for constructing a Siamese network-based content-based music recommendation system that integrates audio features and lyrics. Using a dataset accessible on Kaggle, audio attributes are extracted from the Spotify API and lyrics from the Genius API. In terms of accuracy and user happiness, the suggested solution exceeds already-existing content-based recommendation systems. Unlike collaborative filtering techniques, which tends to propose more mainstream and popular music, this strategy can support up-and-coming and lesser-known musicians by recognizing their distinctive work. Our findings have implications for the creation of more precise and reliable music recommendation systems that consider users' distinct preferences and musical inclinations.