A microscopic view of song repetitiveness

Hannah Yan Han
2 min readApr 1, 2018

How do lyrics repeat themselves and when do the repeated patterns first occur?

I took 3 random songs from last year and show words/phrases in the songs as bubbles sized by the number of repetitions and colored by their first occurrences, with lighting color indicating the words first appeared early in the song. Dark border is applied to identified repetitive phrases or sentences as opposed to individual words.

These songs have all established repeatable patterns early on, then evolved in their own ways. Hello by Adele repeats entire stanza. Havana started the ‘ooh-na-na’ in the first half then brought in more repetition in the second half. Shape of you get rather repetitive in the second half and used a whole lot of the word ‘and’ throughout to connect different actions and sentences.

Interactive version in the making

There are some other song repetitiveness visualizations which are fascinating: SongSim takes an abstract view of overall pattern, and Are Pop Lyrics Getting More Repetitive? by Pudding.cool focuses on distribution of repeated percentage and its change across the decades.

This is #day82 of my #100dayprojects on data science and visual storytelling.

There were two challenges in this project.

  1. Identify words or phrases. While a combination of different n-grams which don’t co-occur by chance are obtainable, the results are kinda messy and still take some some manual curation to sieve out the longest phrases that make semantic sense.
  2. Show change in pattern. Smooth transitional animation which shows the evolving size of bubbles turns out to be challenging here due to the packed circle layout.

The code can be found on my github. If you like it, please share it. Suggestions of better methodology and feedbacks are always welcomed.