Utilizing kALE and Chord Structure in Content-Based Music Recommendation

Abstract

With the popularity of digital entertainment applications such as Pandora, Spotify, Amazon Music, etc., comes the need for sophisticated music recommendation algorithms. Many algorithms depend on collaborative filtering, determining relevant song suggestions for a particular user by analyzing similar tastes in the rest of the user base. This method comes with the “cold start” problem, such that a system does not have enough data on a new user to generate meaningful suggestions. A solution to this problem is a content-based method, utilizing the structure and features of the music itself to determine similar songs rather than user-based data. By using audio analysis tools and performing wavelet analy- sis on songs to extract relevant acoustic features (chord progression), we will define a new measure of similarity between songs. We show that using similarity of chord structure as a basis for content-based music recommendation is a quality metric that can be introduced into existing music recommendation applications.