summary: Researchers have developed a machine learning model that uses neural responses to predict song success with 97% accuracy.
Participants listened to a series of songs while monitoring their neurophysiological responses, generating data that helped machine learning models determine potential hits. Called “neuro-prediction,” this groundbreaking approach helps streaming services efficiently identify popular new songs for playlists.
The researchers believe that, with certain limitations, their method could be applied beyond identifying songs and predicting the hits of movies and TV shows.
Important facts:
- A new “neuro-prediction” approach uses machine learning models applied to neural responses to predict hit songs with 97% accuracy, compared to 50% accuracy for traditional methods.
- Neurophysiological responses to the first minute of the song predicted hits with an 82% success rate. This indicates that the first part of a song plays an important role in determining its popularity.
- Despite limitations such as the relatively small number of songs analyzed and the moderately diverse demographics of the participants, the researchers found their method to be a hit in other entertainment areas. I’m sure it can be predicted.
sauce: frontier
Tens of thousands of songs are released every day. This constant stream of options makes it difficult for streaming services and radio stations to choose which songs to add to their playlists.
These services use human listeners and artificial intelligence to find what resonates with most audiences. However, the accuracy of this approach is only 50%, and he cannot reliably predict whether a song will be a hit.
Now, US researchers were able to predict hit songs with 97% accuracy using comprehensive machine learning techniques applied to brain responses.
“By applying machine learning to neurophysiological data, we were able to identify hit songs with near perfection,” said Paul, a professor at Claremont Graduate University and lead author of the study published in 2006. Zak said. Artificial intelligence frontier.
“It is quite amazing that the neural activity of 33 people can predict whether millions of other people have heard a new song. I have never.”
Machine learning using neurological data
Study participants were equipped with off-the-shelf sensors, listened to a set of 24 songs, and were asked about their preferences and some demographic data. During the experiment, scientists measured participants’ neurophysiological responses to the song.
“The brain signals we collect reflect the activity of brain networks related to mood and energy levels,” says Zack. This allows researchers to predict market outcomes, such as the number of song streams, based on a small amount of data.
This approach is called “neural prediction”. Capture the neural activity of a small number of people to predict population-level impact without measuring the brain activity of hundreds of people.
After data collection, researchers used various statistical approaches to assess the predictive accuracy of neurophysiological variables. This allowed us to directly compare the models. To improve prediction accuracy, we trained an ML model that tested different algorithms to get the best prediction results.
They found that a linear statistical model identified hits with a 69% success rate. After applying machine learning to the collected data, the percentage of correctly identified hits jumped to 97%.
They also applied machine learning to neural responses during the first minute of the song. In this case, hits were correctly identified with his 82% success rate.
“This means that streaming services can more efficiently and easily identify new songs that are likely to hit people’s playlists, making their job easier and pleasing their listeners.” Zach explained.
Replication method
“In the future, wearable neurotechnologies like the one we used in this study will be able to deliver appropriate entertainment to viewers based on their neurophysiology. By giving us just two or three choices instead of being offered to us, it makes it easier and faster to choose the music we enjoy,” Zak said.
Despite his team’s near-perfect prediction results, the researchers noted some limitations. For example, they used a relatively small number of songs in their analysis. Additionally, the demographics of study participants were moderately diverse, but did not include members of specific ethnic or age groups.
Nevertheless, the researchers anticipate that their approach will likely have uses beyond identifying hit songs, due in part to its ease of implementation.
“Our main contribution is methodology. This approach could also be used for hit prediction for many other types of entertainment, such as movies and TV shows,” Zack concluded.
About this music and artificial intelligence research news
author: Deborah Pilchner
sauce: frontier
contact: Deborah Pirchner – Frontiers
image: Image credited to Neuroscience News
Original research: open access.
“Accurate Prediction of Hit Songs Using Neurophysiology and Machine Learning” by Paul J Zak et al. Artificial intelligence frontier
overview
Accurately predict hit songs using neurophysiology and machine learning
Hit songs are notoriously difficult to identify. Traditionally, song elements have been measured from large databases to identify lyrical aspects of hit songs. We took a different methodological approach to measure neurophysiological responses to a set of songs delivered by a streaming music service and identify hits and flops.
We compared several statistical approaches to examine the predictive accuracy of each technique. A linear statistical model using two neural measures identified hits with 69% accuracy. We then created synthetic set data and applied ensemble machine learning to capture the nonlinearities inherent in neural data.
The model classified hits with 97% accuracy. Applying machine learning to his neural responses in the first minute of a song, he found that he was able to correctly classify hits 82% of the time, making his brain identify hits more quickly.
Our results show that applying machine learning to neural data can significantly improve the classification accuracy of hard-to-predict market outcomes.
