Changing Beats Over the Decades: Song Theme Clustering
Data Mining and Wrangling Course Submission
See full report HERE
Executive Summary
"A good song never gets old." This statement encapsulates music across different eras that have captured the hearts of millions of listeners worldwide. Song hits are evident from radio to television and even the digital streaming era, which made us ask "What are the themes present in hit songs?” Our focus is on songs in Spotify, the world’s largest music streaming provider. Spotify curates themed playlists to engage their wide variety of listeners. In this study, hit songs from Spotify All Out Playlists were analyzed. Each collection contains one hundred and fifty hit songs per decade starting from the 1950's all the way to the 2010's. The methodology starts with collection of data using Spotify API where data cleaning was also performed. All audio features were gathered and standardized. The dimensions were reduced by using Principal Component Analysis. The team was then able to identify different clusters within each decade and similar clusters between the decades through KMeans Clustering.
This paper is beneficial for three sectors: Artists, Spotify users, and Spotify, Inc. Our findings can help these sectors understand global musical preference. Artists, musician, and producers are able to ride the waves of popular emotial but upbeat songs, create happy songs, or revive and be inspired by themes of the past such as sentimental dance songs. Using our identified clusters, subscribers are able to discover songs from other decades that are similar to their preferred cluster. Spotify is also able to harness our analysis in clustering to diversify song choices and improve their playlists and catalogue of songs.
Five unique musical themes emerged from our group's analysis: "Easy Pop/Rock", "Senti", "Move", "Chill", and "Alt" . “Easy Pop/Rock” contains songs that are happy, upbeat, and can boost your mood; “Senti” has songs that are generally unhappy and relatable; “Move” hits are great for dancing or parties; “Chill” has songs with relaxing beats, perfect for chill and slow dancing; and “Alt” which has tunes that emotional but upbeat. Among these clusters, “Move” and “Senti” are present most of the decades. “Easy Pop/Rock” songs were most popular before 2000’s while “Alt” emerged at the start of 2000’s. It is noteworthy that the “Chill” cluster emerged when the “Senti” cluster faded and happier songs are present in the majority of each decade. Finally, the “Alt” cluster appeared at the start of 2000’s which ushered in a new wave of song preference that encourages rising above adversity.
For future studies, we recommend that researchers explore more playlists, songs, and features. They can take it a step further by analyzing actual song lyrics instead of just song titles to get a deeper look into themes through clustering. They can also try utilizing more clustering techniques, validation methods, and possibly soft clusters across decades. Next steps include reverse engineering on Spotify song audio features and performing machine learning algorithms.
Contributors
dela Resma, Marvee
Ginez, Zhoya
Inocencio, Ken
Nepomuceno, Colleen
Piquero, Geran
Punzalan, Paolo