The anatomy of a hit song has remained a mystery to researchers looking to dissect what makes some songs soar to the top of the charts and others land with an embarrassing flop.
Thus far, figuring out what qualities may link very different anthems — say, Adele’s “Rolling in the Deep” and LMFAO’s “Sexy and I Know It” — has been more a matter of alchemy than science.
Now European researchers are challenging that notion. Using 50 years’ worth of hit songs on Britain’s top 40 charts, they’ve come up with a computer program that can predict whether a song will catch fire on the airwaves or fizzle out.
They’ve even built an app that allows songwriters and music fans to gauge whether their favorites have hit potential.
Project leader Tijl De Bie, a senior lecturer in artificial intelligence at the University of Bristol in England, discussed the group’s research, which was presented in Spain in December at the MML 2011 4th International Workshop on Machine Learning and Music: Learning from Musical Structure.
You used artificial intelligence to devise an equation that could predict which songs made it to the top of the charts. How does it work?
To predict the hit potential of a given song, we used a computer to quantify how similar it is to previous “hits” and “flops.” Time frame is important: If you’re scoring a song from today, then we will consider the songs in 2011 more important than the songs in the ‘60s.
We represent each song using a set of 23 different features that characterize the audio. Some are very simple features — such as how fast it is, how long the song is — and some are more complex features, such as how energetic the song is, how loud it is, how danceable and how stable the beat is throughout the song. We also took into account the highest rank that songs ever achieved on the chart.
The computer can combine a song’s features in an equation that can be used to score any given song.
We can then evaluate how accurately the computer scored it by seeing how well the song actually did.
Every single week now we’re updating our equation based on how recent releases have done on the chart. So the equation will continue to evolve, because music tastes will evolve as well.
Any good examples of the computer guessing correctly?
Wiley’s “Wearing My Rolex” did well, strongly based on loudness. So that was an expected hit. It went to No. 2 in 2008.
Gnarls Barkley’s “Crazy,” which went to No. 1 in 2006, scored well thanks to its danceability, among other things.
Elvis Presley’s “Suspicious Minds,” which went to No. 2 in 1970, had a fairly simple harmonic movement, which at that time was a good thing if you wanted to score a hit.
Did you notice any trends through the different musical eras, from the ‘60s to now?
Yes, I would say so. Danceability was not important in distinguishing top and bottom songs until the late ‘70s.
But from 1980 it became really important, maybe in relation to the rise of disco, electronic dance music and other later music.
From the late 1980s, the songs at the top became relatively harmonically more complex than songs at the bottom. Before that time, songs on the top tended to be harmonically simpler. That’s quite interesting, because somehow the opposite is true for rhythm.
Nowadays [since the late ‘80s ], simple binary rhythms tend to give a better guarantee to success than complex rhythms, and before it was generally the other way around.
Loudness is increasing, on average, for all songs on the chart. And it is relatively higher near the top compared with the bottom. Of course, this can’t continue — it must plateau or decrease at some point. Recently it seems that this loudness trend is perhaps even reversing, but it’s a bit early to tell that.
Does this give you any hints about where popular music will go next?
If you’re asking if music is going in a single direction, I don’t think that is the case. I think it will continue to evolve in fairly unpredictable ways. Our equation is trying to be resilient to that by learning mostly from the most recent songs and a bit less from the more distant past.
Are there any factors you’d like a version 2.0 to take into account?
Yes, in fact a master’s student I’m supervising is looking into whether the emotional content of the lyrics can be useful — the mood in text, whether a positive or negative emotion is being expressed.
We think an important piece of information would probably be the marketing budget used to promote the song, but unfortunately that’s not public information. The band’s prior popularity could also be factored in.
Does this research have any practical uses?
I think it may help bands without a label, who need a breakthrough, to have a way to find if their songs are hit-like or not.
We just released an app on our scoreahit.com website that allows people to score their own songs or score existing songs that are in the database. We want people to understand exactly what it means that the accuracy is 60% and not 100%: For every 10 predictions our equation makes, it gets it right six times, but it also gets it wrong four times. This is probably no surprise given that it relies on a song’s audio features alone, but it’s a fun thing to try — and it might give some indication of the hit potential.
Is it dangerous to put too much faith in the equation?
Some people have expressed a concern that they think this kind of research shouldn’t be done. They think that if a method like this is used by many people, then all music will become the same.
I disagree with that because I don’t think people can be told what to listen to. And if everyone starts using this method — starts making music lacking any creativity that just tries to satisfy this equation — then I think people will soon start longing for something new, which means the equation will somehow defeat itself.
In fact, I hope it will allow more musicians to enter the music market, which may be beneficial to the quality of music we all listen to.
This interview has been edited for space and clarity.