Computer Algorithm Plays Jukebox Jury

Even the likes of Simon Cowell would admit there’s no way to predict a surefire hit, but researchers in the UK believe they’ve found a formula that gives a pretty good insight.

The research comes from the University of Bristol’s Intelligent Systems Algorithm and was as much about a computer’s ability to learn as it was about music. Staff began by coming up with 23 objective statistics that can measure a song, ranging from duration and tempo to volume and time signature.

They then used a computer to measure these statistics for songs from 1960 onwards, fed in the details of the chart position each song reached, then asked the computer to figure out what weight to give each statistic in a prediction formula.

The researchers then applied the formula to a selection of songs to see whether it could correctly predict each song’s peak performance as being in one of three categories: a hit (meaning a top five ranking), not a hit (meaning it failed to place higher than number 30) or something in between. Overall, the formula predicted correctly roughly 60% of the time.

Rather than compare that figure to the effects of simply guessing at random, the researchers instead compared the predictions to what would have happened if they “cheated” at guessing. This was possible because while the overall proportion of hit vs non hit was equal across the entire testing group of records, the balance between the two varied from year to year — in other words, some years more of the records used for testing were hits than misses, and vice versa.

The researchers thus looked at what effect they’d have got if, for example, they’d known their test group from a particular year was hit-heavy and then simply “predicted” every test record from that year to be a hit. It turned out that the formula was more successful for every year, albeit by varying margins.

The most impressive element of the research is that the computer didn’t simply come up with a set formula, but rather one that took account of the year of the song, with the balance between the component factors changing over time:

From this, the researchers discovered that the trends for hits changed over time: in the 1980s it became more important that a song was slow and “dancable”; in the 1990s a simple binary rhythm such as 4/4 time began to be a big benefit; and not only are songs as a whole getting louder, but hits are generally louder than misses.

The researchers also noted that even with the adjustment over time, the formula had varied success: it worked best in the early 90s and from 2000 on, but did comparatively poorly around 1980. They concluded the late 70s/early 80s may have been a “particularly creative and innovative” era for popular music.

Some examples of songs successfully “predicted” as hits that fitted in neatly with the musical tastes of their times include Elvis Presley with Suspicious Minds in 1970, Simply Red with If You Don’t Know Me By Now in 1989, and Gnarls Barkley with Crazy in 2006.

The researchers also noted several cases where hits defied the formula, usually because cultural events overcame the musical tastes of the time: examples include Do They Know It’s Christmas by Band Aid in 1984; Nessun Dorma in 1990 (popularized in the UK as the theme tune of that year’s FIFA World Cup coverage); Man in the Mirror, re-released after Michael Jackson’s death in 2009; and Trashmen’s Surfin Bird, which hit number three in the charts last year after a Facebook campaign attempted to get it to number one at Christmas and almost succeeded.


Comments are closed.