| dc.description.abstract |
The emotion or mood of music affects the listener in various ways. Nowadays people listen to music on streaming services like Spotify. On streaming services, a playlist is a selection of similar songs customized based on listener preferences. Often, those playlist’s names contain words or phrases that express the emotion of music. In this work, we collected 200 songs from 10 different playlists created by Spotify. It is worth noting that these playlists' names convey a variety of emotions such as sad, crying, discourage, feeling love, tired, missed, chill-out etc, which was used as the emotion label for each song in those playlists. Using audio data collected from Spotify web API, we developed music emotion classification models using various machine learning techniques. Random Forest yielded 0.81 accuracy as the best performance. Moreover, we noticed that Random Forest worked best with only 3 or 4 emotion labels. Later, we also noticed similar results by using K Mean clustering technique. We conclude that based on audio data, those 10 playlists have similar pattern and can be grouped into only 3 or 4 collections. |
|