Science Keeps Trying to Predict Which Songs Will Be Hits

If you’re wondering why so many hit songs seem to sound the same, it might be because they’ve all had a little help from science.

There are a number of software programs that try to predict which songs will be a hit by comparing their characteristics and attributes (key, tempo, melody, loudness, beat and about twenty other features) to hit songs of the past.  These programs are called “learning algorithms,” which means the more music you feed them, the better understanding the programs will have of what constitutes a hit.

You can see why such tools would be embraced by the recording establishment.  Who wouldn’t want to release a sure thing in a marketplace filled with uncertainty?

But human creativity is still very unpredictable.  Deep in a paper presented last week, UK researchers admitted that there was still room for unexpected hits–songs that didn’t fit the formulae.

For example, the programs choked when analyzing music from the very late 70s and early 80s, right around the New Wave era and the birth of MTV.  They also struggled with songs from the latter half of the 90s (maybe they hate nu-metal, too).  Hit songs from performers with unique sounds (U2,  Guns ‘N Roses, Arctic Monkeys, Bloc Party and even Michael Jackson) shouldn’t have been hits.

Watch this video that illustrates how the attributes of popular songs have evolved over the decades. You’ll probably have to see it more than once to absorb it all.

I find all this to be rather fascinating.  To read more about this latest study, go here.  Then skip over to Wired for this story featuring actual equations on predicting hit songs that look like they’re right out of Isaac Asimov’s Foundation series.

And if you still want more, visit a website called Scoreahit.  You might be there for a while.