Fine-Tuning Model Transformer Untuk Pengenalan Emosi Pada Lirik Lagu Berbahasa Indonesia Berdasarkan Model Emosi Russell

Prabasati, Aida Fitrania (2025) Fine-Tuning Model Transformer Untuk Pengenalan Emosi Pada Lirik Lagu Berbahasa Indonesia Berdasarkan Model Emosi Russell. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025211033-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (7MB) | Request a copy

Abstract

Musik dan lirik lagu memiliki peran penting dalam menyampaikan emosi, namun sering terjadi perbedaan interpretasi antara emosi yang ingin disampaikan pencipta lagu dan yang diterima pendengar, khususnya pada lirik lagu pop Indonesia. Untuk mengatasi tantangan ini, penelitian ini menerapkan model transformer dengan kerangka Model Emosi Russell sebagai dasar klasifikasi emosi lirik lagu. Penelitian ini bertujuan mengembangkan dan membandingkan performa tiga model transformer, yaitu IndoBERT, RoBERTa, dan DistilBERT, dalam mengenali emosi dominan pada lirik lagu pop Indonesia. Proses penelitian meliputi preprocessing teks, balancing data menggunakan Random Oversampling, serta fine-tuning model pada dataset lirik lagu pop Indonesia yang telah dianotasi dengan empat label emosi utama (Senang, Sedih, Tenang, Marah) sesuai Model Emosi Russell. Evaluasi dilakukan menggunakan skema 5-fold cross validation dengan metrik accuracy, precision, recall, dan F1-score.
Hasil eksperimen menunjukkan bahwa strategi preprocessing dan balancing data sangat memengaruhi performa model. IndoBERT menunjukkan performa terbaik pada skenario tanpa oversampling dan tanpa full preprocessing, dengan accuracy 68%, precision 63%, recall 59%, dan F1-score 60%. RoBERTa memperoleh hasil optimal pada skenario dengan oversampling tanpa full preprocessing dengan accuracy 65% dan F1-score 57%, sedangkan DistilBERT mencapai accuracy 66% dan F1-score 56%, yaitu 2% lebih rendah untuk accuracy dan 4% lebih rendah untuk F1-score dibandingkan IndoBERT. IndoBERT menjadi model paling unggul untuk klasifikasi emosi lirik lagu pop Indonesia pada skenario tanpa oversampling dan tanpa full preprocessing.
==================================================================================================================================
Music and song lyrics play a significant role in conveying emotions, but differences in interpretation between the emotions intended by songwriters and those perceived by listeners often occur, especially in Indonesian pop song lyrics. To address this challenge, this study applies transformer-based models with the Russell Emotion Model as the framework for emotion classification in song lyrics. The objective of this research is to develop and compare the performance of three transformer models IndoBERT, RoBERTa, and DistilBERT in recognizing the dominant emotion in Indonesian pop song lyrics. The research process includes text preprocessing, data balancing using Random Oversampling, and fine-tuning the models on an annotated dataset of Indonesian pop lyrics with four main emotion labels (Happy, Sad, Calm, Angry) according to the Russell Emotion Model. Evaluation is conducted using a 5-fold cross validation scheme with metrics such as accuracy, precision, recall, and F1-score.
The experimental results show that preprocessing strategies and data balancing significantly affect model performance. IndoBERT achieved the best results in the scenario without oversampling and without full preprocessing, with 68% accuracy, 63% precision, 59% recall, and a 60% F1-score. RoBERTa performed optimally with oversampling and without full preprocessing with accuracy 65% and F1-score 57%, while DistilBERT achieved an accuracy of 66% and an F1-score of 56%, which is 2% lower in accuracy and 4% lower in F1-score compared to IndoBERT. IndoBERT proved to be the most effective model for emotion classification of Indonesian pop song lyrics in the scenario without oversampling and without full preprocessing.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Klasifikasi Emosi, Lirik Lagu, Model Emosi Russell, NLP, Transformer; Emotion Classification, Song Lyrics, Russell Emotion Model, NLP, Transformer
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.9.D343 Data mining. Querying (Computer science) Q Science > QA Mathematics > QA76.9.I52 Information visualization T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Aida Fitrania Prabasati
Date Deposited:	28 Jul 2025 02:42
Last Modified:	28 Jul 2025 02:42
URI:	http://repository.its.ac.id/id/eprint/121912

Actions (login required)

View Item