Future X-Factor and American Idol stars can rest easy in the knowledge scientists have been working tirelessly to take out the guesswork in what makes a hit song. They've developed a software program that can predict chart positions with about 60% accuracy.
Machine learning techniques were used to help the equation learn about the relative importance of all the elements that make up a pop song. The equation was made using public data about songs from the U.K. Top 40 of the last 50 years, and for each week over that timeframe the equation was tested with new releases to see if it could predict where it would land on the charts.
At 60%, the equation gets it more right than wrong but it is still not a completely perfect barometer. What trips it up? The human equation.
Surprisingly, it seems this human equation has less to do with inspired lyrics or riffs, but rather the pressure brought to bear on generating a winner through a social media appeal or through marketing and PR. Researchers pointed to a case in 2010 where Surfin' Bird by The Trashmen landed at number three in the U.K. charts thanks to a web campaign persuading people to buy it over a competing song.
Dr. Tijl De Bie, a senior lecturer in artificial intelligence at Bristol, who headed the research team did point out to the BBC, "At every moment in time the equation can be different because we only took into account past data."
He was also very careful to point out predicting a hit was no guide to whether a song was actually good or worth listening to. "It's not a value judgment," he said, adding it is merely a social barometer for measuring what people were more likely to buy.
Dr. De Bie said of the research, "The goal was to find out if we could come up with an equation that distinguishes between a hit and something that dangles at the bottom of the charts."
Okay, so now I get why they did the research — if it's about what people might buy rather than the inherent quality of the tunes then I can see this being very valuable to the music industry.
From a strictly anthropological point of view I will admit to being interested in the cultural trends that might be reflected by the changes in the hits over the years. For example, the analysis showed that music had become easier to dance to and louder over time.
But the science side of me still keeps asking questions. Exactly how does the equation measure these two qualities — especially danceability? What would Dick Clark say about all this?
I'll admit to feeling a little used up inside when I realize a computer can predict a large portion of whom and what will be successful in the music business. It seems to takes away the reward for inspiration or sheer genius and assigns it to the fringes of the business.
Then again, when I recall "The Macarena" and "I'm Too Sexy" were hits I shouldn't be so surprised.