| Sumario: | This dissertation explores the enhancement of music recommendation systems by integrating semantic similarity in Spanish song lyrics, utilizing advancements in machine learning and natural language processing (NLP), including both supervised and unsupervised approaches. It addresses the gap in current recommendation practices, which often overlook the rich semantic content of lyrics, despite their potential to significantly personalize music recommendations. Through theoretical insights into word embeddings and transfer learning, the development of the LyricSIM dataset for assessing lyric similarity, and empirical evaluations of models designed to distinguish between similar and non-similar song pairs, this research proposes a novel, lyrics-driven approach to music recommendation. Focused on the Spanish-speaking market, where Latin music is prevalent, this study contributes to the field by demonstrating how NLP technologies can refine music recommendations, addressing challenges like the cold start problem and enhancing the diversity of music recommendations, thereby offering a more personalized and engaging user experience in the streaming era.
|
|---|
