A Computational Model for Predicting Music Popularity: A Psychophysiological Study

Abstract

Abstract Hit Song Science (HSS) is pivotal in hit song identification. Music producers and composers often rely on new musical pieces to succeed with no prior knowledge. Hit song science refers to the ability of identifying potential hits relying on various information such as no of streams, no of album sales, likes and listener responses. Among various techniques, Electroencephalogram (EEG) has gained significant attention in the research community as a potential source for gathering lister responses. This research focuses on development of a computation model for predicting hit songs by analyzing psychophysiological responses. This research took a methodological approach in measuring psychophysiological responses to a set of songs provided by a music chart. The identified musical pieces were divided into hits and flops. Their responses were utilized in developing new deep learning models using Convolutional Neural Networks (CNN) and Long Short Term Memory (LSTM) techniques for identifying hits and flops using listener responses. We evaluated the models on two tasks: hit song classification and musical chart ranking prediction. The CNN model trained in Week 13 achieved the highest classification accuracy at 65.43%. For ranking prediction, the best performance was observed with the Week 10 model, which achieved a Mean Squared Error (MSE) of 179.03. These results highlight the potential of deep learning techniques in leveraging psychophysiological signals to improve the accuracy of hit song prediction. The dataset acquired during the experiment is made available to the public, and researchers are encouraged to use it to test their own hit music identification techniques.