Three US researchers have created a new machine learning technique to predict which songs will become popular. Whereas previous attempts could only manage to identify hit songs with 50 percent accuracy at most, this new technique boasts a 97 percent predictive accuracy.
Instead of analyzing the lyrics, measuring social media buzz, and other methods used in other forecasting attempts, researchers from Claremont Graduate University used what they call neuroforecasting, also known as the “brain as predictor” approach.
Essentially, this method can predict population outcomes by analyzing the brain activity data of a small group of participants. In the study, only thirty-three participants aged 18 to 57 were fitted with Rhythm + PPG cardiac sensors to check their neurophysiological responses as they listened to 24 recent songs. Surveys were also handed out to gather demographic information.
See also: How to spot AI-generated images
By using a traditional linear logistic regression model, the researchers were able to distinguish which songs are hits or duds with a 69 percent success rate. But by adding neural data into the mix, the researchers were able to increase the predictive accuracy of their Artificial Intelligence (AI) to 97 percent. The research also emphasized their use of an ensemble model as opposed to a single algorithm to capture the complex and non-linear relationships inherent in neurophysiologic data, which further underscored the efficacy of their approach.
The researchers believe that these findings could enhance the effectiveness of features of streaming services for offering personalized soundtracks and discovering new music. By just listening to the first minute of a new song, consumers through their neurologic immersion will get better curated recommendations.
Earlier this year, Spotify launched its AI-powered feature called DJ, a virtual assistant to offer recommendations based on user preferences as well as offer commentary on upcoming songs and artists.
Source: Frontiers