Modelagem de tópicos e evolução temporal: uma análise das músicas de maior audiência no Brasil

Abstract

Given the historical importance of radio in Brazilian music broadcasting, this study analyzes chart-topping lyrics to track changes in musical preferences over time. A substantial dataset was gathered via web scraping and processed using Natural Language Processing (NLP) techniques to transform text into structured numerical embeddings. The research focuses on clustering compositions into predominant topics using probabilistic models, specifically the infinite mixture model under a Bayesian approach. The results reveal thematic patterns in the compositions, highlighting semantic similarities that permeate the different genres and periods within the analyzed dataset.