Home SECURITY Brain Signals Help Scientists Find Hit Songs Using Machine Learning

Brain Signals Help Scientists Find Hit Songs Using Machine Learning

Brain Signals Help Scientists Find Hit Songs Using Machine Learning


Brain Signals Help Scientists Find Hit Songs Using Machine Learning

AI can predict with 97% accuracy whether a song will become a hit.

Among the tens of thousands of new songs that are released every day in the world, how to choose those that will appeal to millions of listeners? Streaming services and radio stations face this problem all the time. They bring in both human experts and artificial intelligence to look for potential hits. But these methods do not always work and have an accuracy of about 50%.

US scientists have found a new way to solve the problem. The researchers used a complex machine learning technique applied to the brain responses of people who listened to different songs. They were able to predict hit songs with 97% accuracy. The study was published in Frontiers in Artificial Intelligence.

The experiment involved 33 people. They were asked to listen to 24 songs that were preselected by the streaming service as hits or flops. After listening to each song, participants answered questions about their preferences and some demographics. During the experiment, the researchers measured participants’ neurophysiological responses to songs using sensors that recorded the activity of the brain circuit associated with mood and energy levels.

“By applying machine learning to neurophysiological data, we were able to identify hit songs almost perfectly,” said Paul Zack, professor at Claremont University and senior author of the study. “The way the neural activity of 33 people can predict what millions of others will listen to is pretty amazing. No other method has shown such accuracy before.”

The scientists compared several statistical approaches to assess the predictive accuracy of neurophysiological variables. They found that a linear statistical model using two neural measurements determined hits with an accuracy of 69%. They then created a synthetic dataset and applied ensemble machine learning to account for non-linearities in the neural data. This model classified hit songs with 97% accuracy. In addition, they applied machine learning to neural responses to the first minute of songs. In this case, the hits were correctly determined with an accuracy of 82%.

“This means that streaming services can easily identify new songs that are likely to be hits on people’s playlists, making things easier and delighting listeners,” Zack said. He also noted that the technique could have other applications in neuromarketing and consumer neuroscience.


Source link



Please enter your comment!
Please enter your name here