Αυτόματη ταξινόμηση μουσικολογικού είδους με βάση τους στίχους
Lyrics-based music genre classification
View/
Keywords
Ταξινόμηση μουσικών ειδών ; Επεξεργασία φυσικής γλώσσας ; NLP ; BERT ; Ταξινόμηση κειμένων ; Μηχανική μάθηση ; Μοντέλα μετασχηματιστών ; Προεπεξεργασία δεδομένων ; Βελτιστοποίηση μοντέλου ; Νευρωνικά δίκτυαAbstract
This thesis investigates the development and optimization of a BERT-based model for the automatic classification of music genres. Through extensive data preprocessing and model fine-tuning, we addressed the challenges posed by imbalanced datasets and common word overlaps across genres. Initial trials with five genres yielded suboptimal results, leading to adjustments in genre selection to improve model performance. Subsequent experiments with fewer genres achieved notable improvements, with the final model accurately classifying three genres with an impressive 93% accuracy. This research not only highlights the efficacy of transformer models in text classification tasks but also provides insights into optimizing machine learning workflows for practical applications in music genre classification.