Stefan Milosevic

This study pioneers the intersection of Hit Song Science, Graph Neural Networks (GNNs), and neuroscience to forecast the popularity of songs within the Italian Pop genre on Spotify using EEG signals. Our aim is to develop a graph representation learning framework capable of predicting a song’s popularity index—a measure derived from the streaming count normalized over the song’s duration on the platform. We utilize electroencephalogram (EEG) recordings from 27 individuals (13 males and 14… Show more

This study pioneers the intersection of Hit Song Science, Graph Neural Networks (GNNs), and neuroscience to forecast the popularity of songs within the Italian Pop genre on Spotify using EEG signals. Our aim is to develop a graph representation learning framework capable of predicting a song’s popularity index—a measure derived from the streaming count normalized over the song’s duration on the platform. We utilize electroencephalogram (EEG) recordings from 27 individuals (13 males and 14 females, aged 20-30 years), who listened to 20 tracks each, yielding 540 EEG samples. These samples were recorded using a g.Nautilus device with 32 channels, capturing the brain’s electrical activity as subjects were exposed to music. This innovative approach incorporates not only the temporal dimension of EEG data but also integrates Spotify’s audio features to enrich our prediction model. We propose a novel application of Temporal Graph Neural Networks (Temporal GNNs) to analyze the complex, dynamic interactions within the brain as it responds to musical stimuli. By providing a suitable graph connectivity for each EEG recording and developing a predictive model, we aim to bridge subjective musical experience with objective market success. Our work is positioned at the confluence of computational neuroscience, deep learning, and musicology, promising not only to advance the field of neuromarketing but also to contribute significantly to our understanding of the neural correlates of musical preference and its impact on consumer behavior. This research underlines the potential of machine learning and neural analysis in uncovering the intricacies of human auditory perception and its implications for the music industry. Show less

Worked with chest-Xray-pneumonia datasets.

Worked collaboratively with Oncologists to identify the needs of hospital’s patients through a mobile application.