AI 'Hit Song Predictor' Developed to Foresee Blockbuster Hits Like 'Cupid'...

Neurophysiology Data Machine Learning
Hit Prediction Accuracy Reaches 97%

American researchers have developed a hit song predictor using machine learning.

On the 20th, a research team led by Professor Paul Jack of Claremont Graduate University in the United States announced in the scientific journal Frontiers in Artificial Intelligence that they successfully applied machine learning to brain response data to songs, predicting hit potential with 97% accuracy.

Music streaming services and broadcasters have used various methods such as listener sampling surveys and AI activities to find songs that resonate with people among the numerous songs released daily. However, the prediction accuracy has been known to be around 50%.

The research team attached sensors to the heads of 33 experimental participants and played 24 songs provided by streaming services while measuring the neurophysiological responses of their brains. Along with this, they surveyed preferences for each song and demographic information through questionnaires.

After collecting brain activity data from the participants, the researchers evaluated the hit prediction accuracy of neurophysiological variables and trained a machine learning model to achieve the highest accuracy.

Through this, they identified a linear statistical model that could select hit songs with 69% accuracy using two neuro-measurements, and succeeded in increasing the hit song prediction accuracy to 97% by applying machine learning to the collected neurophysiological data.

Additionally, this AI hit song predictor was found to predict hit songs with 82% accuracy using only the neural response to the first minute of a song.

"Applicable to movies and TV programs... Will provide entertainment suitable for people"

Professor Jack explained, "Brain signals collected by sensors reflect the activity of brain networks related to the mood and energy levels of the experimental participants," adding, "Based on a few data points, it is possible to predict market outcomes such as the number of streams of a song."

He emphasized, "There has been no case so far of selecting hit songs with this level of accuracy using neurophysiological data," and added, "It is remarkable that the neural activity data of 33 people can predict the responses of millions to a song."

He also stated, "This result means that streaming services can efficiently identify new songs likely to become hits, making their work easier and providing enjoyment to listeners," and "In the future, if wearable neuroscience technology like that used in this study becomes widespread, it will be possible to provide entertainment suitable for people based on neurophysiology."

Professor Jack also added, "This approach is easy to implement and could be used in other fields such as movies and TV programs."

Meanwhile, the research team noted that although this system has very high prediction accuracy, the study has limitations such as the relatively small number of songs used in the experiment and the lack of certain races and age groups among the participants.

This content was produced with the assistance of AI translation services.