Skip to content

jakhac/song-popularity

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

71 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Prediction of a Song's Popularity

Data science approach to predict the popularity of a song using ensemble learning methods for musical, lyrical and artist features. This repository contains the code associated with our term paper.

Machine learning approach

Abstract

We predicted song popularity using musical and artist features from Spotify data as well as self-computed lyrical measurements. For each set of features, we trained binary and multiclass classifiers and found the artist-only model to have better results than the music-only and lyrics-only models. We implemented two ensemble approaches and achieved the overall best result using a random forest classifier with an accuracy of 82% for binary and 65% for multiclass classification. The most decisive features ordered by relevance are artist follower count, artist popularity and release year. We found out that only 42% of popular songs would also be popular if released by an unpopular artist. Consequently, song popularity largely depends on the popularity of an artist.

Code

Please note that the code is not fully executable without a SQL database populated with the Spotify dataset from Kaggle (Nov 21: not available anymore, use a similar dataset or the Spotify API directly) and the lyrics corresponding to the songs. We do not publish the database, however its structure can be re-created by running the db_setup.py script and then using the scripts in the dataset folder to populate it. To work with lyrics, a Genius API key will be required.

Contributors

Kevin Katzkowski - GitHub
Jakob Hackstein - GitHub
Luca D'Agostino

About

Tool for predicting a song's popularity

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published