Every day, a large number of songs are released, presenting a challenge for streaming services and radio stations to select which ones to add to playlists. To overcome this difficulty and find songs that will appeal to a wide audience, these services have relied on human listeners and artificial intelligence. However, this approach has only achieved a 50% accuracy rate in predicting whether a song will become a hit.
Researchers in the United States have now utilized a comprehensive machine learning technique combined with brain responses to predict hit songs with an impressive 97% accuracy.
“By applying machine learning to neurophysiologic data, we could almost perfectly identify hit songs,” said Paul Zak, a professor at Claremont Graduate University and the senior author of the study published in Frontiers in Artificial Intelligence. He further added, “That the neural activity of 33 people can predict if millions of others listened to new songs is quite amazing. Nothing close to this accuracy has ever been shown before.”
In the study, participants were equipped with off-the-shelf sensors and listened to a set of 24 songs while providing feedback on their preferences and some demographic information. The researchers measured the participants’ brain responses to the songs during the experiment. Zak explained, “The brain signals we’ve collected reflect activity of a brain network associated with mood and energy levels.” This allowed the researchers to predict market outcomes, such as the number of song streams, based on data from a small group.
This innovative approach is referred to as “neuroforecasting.” It captures the neural activity of a small group of people to predict effects at a population level, eliminating the need to measure the brain activity of hundreds of individuals.
After collecting the data, the researchers employed various statistical approaches to evaluate the accuracy of their neurophysiological variables in predicting hits. They trained a machine learning model using different algorithms to improve the predictive accuracy.
The results showed that a linear statistical model correctly identified hit songs at a rate of 69%. However, when machine learning was applied to the collected data, the accuracy of hit song identification soared to 97%. Additionally, applying machine learning to the neural responses within the first minute of the songs resulted in a success rate of 82%.
Zak explained the significance of these findings, stating, “This means that streaming services can readily identify new songs that are likely to be hits for people’s playlists more efficiently, making the streaming services’ jobs easier and delighting listeners.”
Looking to the future, Zak envisioned a scenario where wearable neuroscience technologies, like the ones used in the study, become widespread. This would enable tailored entertainment experiences based on individuals’ neurophysiology. Instead of being overwhelmed with countless options, people could be presented with just a few choices, making it quicker and easier for them to select music they would enjoy.
While the researchers achieved near-perfect prediction results, they acknowledged certain limitations. They used a relatively small number of songs in their analysis, and the study participants represented a moderately diverse range of demographics but did not include members from certain ethnic and age groups.
Nonetheless, the researchers believe that their methodology can be applied beyond hit song identification due to its straightforward implementation. Zak concluded, “Our key contribution is the methodology. It is likely that this approach can be used to predict hits for many other kinds of entertainment too, including movies and TV shows.”
