Hit song classification with spotify audio features of Billboard list songs
Lindholm, Juuso (2022)
2022
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2022053012986
https://urn.fi/URN:NBN:fi:amk-2022053012986
Tiivistelmä
In this work an experiment was made to predict hit and non-hit songs based on audio features provided by Spotify. Four different models were compared with the dataset.
The billboard data was first collected from Kaggle. The collected data did not include a Spotify id connecting the song from Kaggle to a song in Spotify collections. In order to retrieve the audio features of a song, that will be used as input to the models, a mapping had to be implemented. The implementation was done by retrieving songs from Spotify search API and matching the name of song and artist to the ones used to make the query. During the matching process, some words in the songs were excluded that were referring to different versions of the original song.
After retrieving the track information, the audio features were collected from Spotify audio features API.
After storing the hit song features, the non-hit songs were collected by using the Spotify al-bum id in the song information. The non-hit songs were randomly sampled from the album and a check to verify the song not being the same as what was used to make the query. Finally, the audio features for the non-hit songs were collected the same way as for the hit songs.
The data was processed, transformed and hyperparameters were searched.
The billboard data was first collected from Kaggle. The collected data did not include a Spotify id connecting the song from Kaggle to a song in Spotify collections. In order to retrieve the audio features of a song, that will be used as input to the models, a mapping had to be implemented. The implementation was done by retrieving songs from Spotify search API and matching the name of song and artist to the ones used to make the query. During the matching process, some words in the songs were excluded that were referring to different versions of the original song.
After retrieving the track information, the audio features were collected from Spotify audio features API.
After storing the hit song features, the non-hit songs were collected by using the Spotify al-bum id in the song information. The non-hit songs were randomly sampled from the album and a check to verify the song not being the same as what was used to make the query. Finally, the audio features for the non-hit songs were collected the same way as for the hit songs.
The data was processed, transformed and hyperparameters were searched.