17 December 2022 in Scientific Reports
The emergence of streaming services, e.g., Spotify, has changed the way people listen to music and the way professional musicians achieve fame and success. Classical music has been the backbone of Western media for a long time, but Spotify has introduced the public to a much wider variety of music, also opening a new venue for professional musicians to gain exposure. In this paper, we use open-source data from Spotify and Musicbrainz databases to construct collaboration-based and genre-based networks. We call genres defined in these databases primary genres. Our goal is to find the correlation between various features of each professional musician, the current stage of their career, and the level of their success in the music field. We build regression models using XGBoost to first analyze correlation between features provided by Spotify. We then analyze the correlation between the digital music world of Spotify and the more traditional world of Billboard charts. We find that within certain bounds, machine learning techniques such as decision tree classifiers and Q-based models perform quite well on predicting success of professional musicians from the data on their early careers. We also find features that are highly predictive of their success. The most prominent among them are the musicians’ collaboration counts and the span of their career. Our findings also show that classical musicians are still very centrally placed in the general, genre-agnostic network of musicians. Using these models and success metrics, aspiring professional musicians can check if their chances for career success could be improved by increasing their specific success measures in both Spotify and Billboard charts.
24 March 2024 in Proceedings of the AAAI Conference on Artificial Intelligence
Data distillation is a technique of reducing a large dataset into a smaller dataset. The smaller dataset can then be used to train a model which can perform comparably to a model trained on the full dataset. Past works have examined this approach for image datasets, focusing on neural networks as target models. However, tabular datasets pose new challenges not seen in images. A sample in tabular dataset is a one dimensional vector unlike the two (or three) dimensional pixel grid of images, and Non-NN models such as XGBoost can often outperform neural network (NN) based models. Our contribution in this work is two-fold: 1) We show in our work that data distillation methods from images do not translate directly to tabular data; 2) We propose a new distillation method that consistently outperforms the baseline for multiple different models, including non-NN models such as XGBoost.