Music, Data, and Discovery: Innovative Approaches to Understanding Musical Trends and Recommendation

Statistical modeling of popular music presents a unique challenge due to the complexity of song structures, which cannot be easily analyzed using conventional statistical tools. However, recent advances in data science have shown that converting non-standard data objects into real vector-valued embeddings enables meaningful statistical analysis. In this talk, we demonstrate an approach based on logistic principal component analysis (logistic PCA; Landgraf and Lee, 2015) to construct embeddings from global song features, allowing for standard multivariate analysis. We apply this method to a corpus of Lennon and McCartney songs from 1962–1966, using embeddings derived from chords, pitch transitions, and melodic contours. Our analysis explores how these song embeddings cluster by Beatles album, how songwriting styles evolved over time, and whether Lennon and McCartney's compositions exhibited convergence or divergence. This embedding-based approach offers a powerful framework for statistically examining musical structure and stylistic development in popular music. 

In light of the central role that online recommendations have taken in people's lives, and given the limitations and wide room for improvement for music recommendation systems currently in place, this work proposes a novel content-based approach for generating music recommendations solely based on the detection of semantic and sonic connections between musical artists and the construction and programmatic analysis of a latent directed graph embedded in such connections. The core idea behind this approach is that traversing such a graph essentially emulates the humanly unfeasible but highly intuitive process of reading through an arbitrary-length sequence of high-quality music reviews, using any given artist as a starting point and hoping from one review to the next according to the list of different artists found in each read, and hierarchically prioritizing the queue by semantic and sonic similarity, two metrics that, in tandem, can be plausibly interpreted as musical influence in this context. First, I describe the background and motivation for developing this framework, and explain its novelty and main benefits. I then explain how connections between artists are drawn, how the graph is built, and how this the graph can be enhanced with sonic attributes and used programmatically to produce music recommendations in the form of optimal sequences of artists and/or songs. Finally, I presents an application of this framework using reviews from several reputable online music publications, presents some of the resulting graph properties, which prove to be highly desirable in terms of connectivity, and presents a publicly available interactive tool built on top of that graph.I conclude by summarizing the main strengths of this framework, discussing its limitations, and pointing to some potential further enhancements. 

Popular music genre preferences can be measured by consumer sales, listening habits, and critics' opinions. We analyze trends in genre preferences from 1974 through 2018 presented in annual Billboard Hot 100 charts and annual Village Voice Pazz & Jop critics' polls. We model yearly counts of appearances in these lists for eight music genres with two multinomial logit models, using various demographic, social, and industry variables as predictors. Since the counts are correlated over time, we use a partial likelihood approach to fit the models. Our models provide strong fits to the observed genre proportions and illuminate trends in the popularity of genres over the sampled years, such as the rise of country music and the decline of rock music in consumer preferences, and the rise of rap/hip-hop in popularity among both consumers and critics. We forecast the genre proportions (for consumers and critics) for 2019 using fitted multinomial probabilities constructed from forecasts of 2019 predictor values and compare our Hot 100 forecasts to observed 2019 Hot 100 proportions. We model over time the association between consumer and critics' preferences using Cramér's measure of association between nominal variables and forecast future trends for this association.  We also present some recent data since 2018 to revisit genre trends in the years since our data set was collected.