I Said it First: Topological Analysis of Lyrical Influence Networks

We present an analysis of musical influence using intact lyrics of over 550,000 songs, extending existing research on lyrics through a novel approach using directed networks. We form networks of lyrical influence over time at the level of three-word phrases, weighted by tf-idf. An edge reduction analysis of strongly connected components suggests highly central artist, songwriter, and genre network topologies. Visualizations of the genre network based on multidimensional scaling confirm network centrality and provide insight into the most influential genres at the heart of the network. Next, we present metrics for influence and self-referential behavior, examining their interactions with network centrality and with the genre diversity of songwriters. Here, we uncover a negative correlation between songwriters’ genre diversity and the robustness of their connections. By examining trends among the data for top genres, songwriters, and artists, we address questions related to clustering, influence, and isolation of nodes in the networks. We conclude by discussing promising future applications of lyrical influence networks in music information retrieval research. The networks constructed in this study are made publicly available for research purposes.