A Novel Feature-Based Music Recommendation System Considering the Uniqueness of Musical Items

12 Pages Posted: 24 Jan 2023 Last revised: 13 Apr 2023

See all articles by Siphesihle Ndhlovu

Siphesihle Ndhlovu

University of the Witwatersrand

Ritesh Ajoodha

University of the Witwatersrand

Date Written: November 23, 2022

Abstract

Some music recommendation systems do not consider the uniqueness of song items well enough to be able to classify these by the general mood and feel they may give off. The purpose of this report is to thus create a system that attempts to solve this problem and create a model to categorise each musical item by mood using their content-based features. The model of choice is the Gaussian Maximization Model since it allows the creation of inferences modelled in a computationally efficient way. The dataset to be explored is the publicly available Spotify dataset which comes with pre-processed content-based features of different songs from the era 1920- 2020. Once this is done, I will then compare the different silhouette scores and Bayesian Inference Criterion (BIC) score obtained by this model with the score obtained by KMeans clustering algorithm which serves as the benchmark model we hope to improve on.

Keywords: Music recommendation systems, content-based, mood-based clustering, Gaussian mixture models, silhouette score

Suggested Citation

Ndhlovu, Siphesihle and Ajoodha, Ritesh, A Novel Feature-Based Music Recommendation System Considering the Uniqueness of Musical Items (November 23, 2022). Available at SSRN: https://ssrn.com/abstract=4332853 or http://dx.doi.org/10.2139/ssrn.4332853

Siphesihle Ndhlovu

University of the Witwatersrand

Ritesh Ajoodha (Contact Author)

University of the Witwatersrand ( email )

Do you have a job opening that you would like to promote on SSRN?

Paper statistics

Downloads
40
Abstract Views
208
PlumX Metrics