®®®® SIIA Público

Título del libro:
Título del capítulo: Unsupervised Emotion Analysis in Popular Music Lyrics in Mexico Using a BERT-Based Model

Autores UNAM:
MARISOL FLORES GARRIDO;
Autores externos:

Idioma:

Año de publicación:
2025
Palabras clave:

Behavioral research; Computer music; K-means clustering; Large datasets; Psychology computing; BERT; Clusterings; Emotion analysis; Emotion detection; Emotional analysis; Emotional patterns; Me-xico; Popular music; Song lyric; Unsupervised analysis; Emotion Recognition


Resumen:

This paper presents an unsupervised analysis of emotional patterns in popular lyrics in Mexico across seven decades. We introduce a corpus of 629 number-one hits, compiled from established sources starting in 1950. Our analysis includes lexical diversity, predominant emotion detection, and song characterization based on emotional profiles. For emotion detection, we use RoBERTuito, a BERT-based model trained for emotion classification in Spanish, to compute emotional scores for individual phrases. These scores are aggregated into statistical feature vectors representing each song across six emotions. Applying k-Means clustering, we identify three archetypal song categories and analyze their distribution over time. Additionally, we explore sequences of predominant emotions within songs, uncovering structural patterns in emotional expression. Despite challenges posed by the poetic and ambiguous nature of lyrics, our scalable computational framework advances digital humanities through a unique dataset, replicable emotion-analysis methodology, and the first large-scale data-driven study of popular lyrics in Mexico. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.


Entidades citadas de la UNAM: