An Interactive Visualization of Billboard Song Clusters
By Team M&L from SNU



When our team first met, we realized we were all music enthusiasts. So we tried searching for music related data, and found a dataset with numerical audio features (provided by Spotify) on songs from the Billboard (1999 - 2019). You can find the data here.

With the dataset, we wanted to explore topics like trends and clustering. Thus, we used the audio feature values to form unique vectors for each song and performed k-means clustering to produce 5 clusters. Since we wondered if people with different tastes would fit in different clusters, we used the data to create a survey that leads to visualizations of a billboard song cluster that best defines a person.

Before the survery, a circular bar chart of the no.1 billboard songs is shown for an understanding of the features, their relations, and possible trends. The bar lengths represent each song’s feature value ratio to the max feature value in the dataset, and you can check the exact ratios by hovering. Try to pick a song you know, imagine the features the song may have, and compare with the actual data! You can choose which feature to see and how to sort it on the top. Here is an explanation of each feature.

Acousticness: Whether the track is acoustic
Danceability: How suitable a track is for dancing
Energy: Perceptual measure of intensity and activity
Key: Estimated overall key of the track
Tempo: Overall estimated tempo of a track in beats per minute (BPM)
Valence: Musical positiveness conveyed by a track

To discover relations between features, try sorting features with other features
(ex. Whether more acoustics songs are slower).

To discover trends, try to sort each feature by year
(ex. If overall valence changes over time).

Now, click the button below to discover your music cluster!
P.S. Pay attention to the background colors as you answer the questions 😊

"Can music be represented with numbers?"