A Music Recommendation System Based on Song Lyrics

Abstract

The goal of this paper is to produce the most effective music recommendation relying solely on the lyrics of said songs. It uses a dataset of song lyrics scraped from the music lyrics website MetroLyrics.com. The recommendations are produced using Euclidean and Cosine distance along with Kullback-Leibler divergence across lyrics representations of bag-of-words, Term Frequency Inverse-Document Frequency (TF-IDF), and Latent Dirichlet Allocation. Ultimately it describes that Cosine distance on TF-IDF representations of song lyrics provide a great tradeoff of ease of implementation and successful recommendations.