This repository hosts code and data for our project for Stanford's CS224W (Social Information and Network Analysis).
Marco Alban, Vivek Choksi, Stephanie Tsai
In our project, we address the overarching question: can we use audio data to identify and characterize musical influence relationships? Using human-annotated data about popular musicians from allmusic.com, we construct a musical influence graph wherein a directed edge between two artist nodes signifies a musical influence relationship. We perform link prediction on this graph by training a classifier with input features based on graph structure as well as song audio content. The results demonstrate that features based on song audio content have predictive power, and our highest-scoring feature combina- tion achieves an area under the receiver operating characteristic (ROC) curve of 0.867. These findings show that audio features of songs can, to some extent, identify musical influence relationships between artists, and this finding affirms the promise of data-driven analyses of the progression of musical creativity.