Text representations for lyric-based identification of musical subgenres

The advancement of techniques and computational tools for data mining has been boosting the music market with applications focused on user experience. These techniques explore musical data looking for patterns and trends that can guide business strategies. One of the key steps in these applications is the vector representation of the original text. This work approaches textual representation techniques applied to the problem of classifying musical sub-genres, a gap in the literature in musical information retrieval, whose complexity lies in the difficult identification of the separation boundary between the sub-classes of the same genre since both carry several features in common. For this, exhaustive experiments were carried out aiming to find the best combination between classifier and textual representation models. The results showed enriched Bag-of-Words (BoW) with the SVM and Logistic Regression algorithms obtained better results than embeddings models and deep neural networks. The conclusions obtained could guide future studies for classifying texts whose separability surfaces are subtle and challenging.