ANALISIS SENTIMEN KOMENTAR YOUTUBE PERTANDINGAN PORTUGAL MELAWAN KROASIA MENGGUNAKAN TRANSFER LEARNING INDOBERT

Authors

  • Syahid Ahmad Yasin Universitas Bina Sarana Informatika
  • Josua Eriek Kurniawan Universitas Bina Sarana Informatika
  • Raditya Fernanda Putra Universitas Bina Sarana Informatika
  • Yumi Novita Dewi Universitas Bina Sarana Informatika
  • Waeisul Bismi Universitas Bina Sarana Informatika
  • Suharyanto Universitas Bina Sarana Informatika

DOI:

https://doi.org/10.35316/jimi.v11i2.10816

Keywords:

sentiment analysis, Transfer learning, Youtube comments, Video Asisstant Referee, IndoBERT

Abstract

Many social media users become furious when refereeing decisions occur in football matches. The highlight of the Portugal vs Croatia match in July-October 2026 was uploaded to TVRI Nasional YouTube channel and received thousands of comments from users criticizing the Video Assistant Referee (VAR) decisions. Unfortunately, in the field of natural language processing (NLP), there is still little research on how to link social media users’ anger to specific events that occurred on the pitch. In this research, we used 5,460 raw comments from the social media and filtered them using the Indonesian word ratio (Rid ≥ 0.40) and the football term dictionary, resulting in 3,273 clean texts (stratified 80:20 split, 262 lines for validation). We used the IndoBERT architecture, which uses the [CLS] token without dropout. On 655 test data (seed 42), the model achieved 75.42% accuracy and 72.29% Macro-F1 (ROC-AUC 87.88%; PR-AUC 76.21%) with a reading speed of 5.860 ms per text on GPU. The average of three seeds was 75.47 ± 0.23% accuracy and 72.37 ± 0.13% Macro-F1. This model outperformed four classic TF-IDF-based methods, i.e., SVM (67.63%), Logistic Regression (66.26%), Random Forest (65.34%), and Naive Bayes (61.83%) (Wilcoxon test p = 0.004; paired t-test p = 0.001). Negative-toned sentences were highly recognizable to the model with an F1-score of 82.05%.On the other hand, the model’s performance on guessing the neutral class dropped to 62.19% due to the fact that social media users’ question sentences often use words identical to those used in complaints. Using logistic regression, we found that VAR offside line debates increased the chance of negative comments by 3.44 times (73.91% negative) and penalty checks increased the chance by 2.25 times (64.93% negative). Finally, we would like to note that the training corpus used in this research still employs automatic labeling with agreement with humans of only 34.67% (Kappa 0.02) and 36.75% on consensus sample (Kappa 0.05).

Downloads

Download data is not yet available.

References

[1] A. G. Ghifari, G. Y. Ananada, dan K. Purwandari, "A Comparative Sentiment Analysis of Public Opinion on Indonesia's National Football Coach Using CRNN and SVM," Procedia Computer Science, vol. 269, hal. 1485-1493, 2025, doi: 10.1016/j.procs.2025.09.090.

[2] G. Sumantri dan B. S. H. Marwoto, "Analisis Sentimen di Twitter Terkait Tim Nasional Sepak Bola Indonesia Menggunakan Metode Support Vector Machine," Jurnal Kajian dan Terapan Matematika, vol. 10, no. 2, hal. 96-104, 2024, doi: 10.21831/jktm.v10i2.19561.

[3] D. A. Prabowo dan Sudianto, "The sentiment analysis of Indonesian football on Twitter by using K nearest neighbors and random forest," JSAI (Journal Scientific and Applied Informatics), vol. 6, no. 2, hal. 247-257, 2023, doi: 10.36085/jsai.v6i2.5337.

[4] A. Khan, D. Majumdar, dan B. Mondal, "Sentiment analysis of emoji fused reviews using machine learning and BERT," Scientific Reports, vol. 15, no. 1, Art. no. 7538, 2025, doi: 10.1038/s41598-025-92286-0.

[5] A. F. Hidayatullah, R. A. Apong, D. T. C. Lai, dan A. Qazi, "Pre-trained language model for code-mixed text in Indonesian, Javanese, and English using transformer," Social Network Analysis and Mining, vol. 15, no. 1, Art. no. 30, 2025, doi: 10.1007/s13278-025-01444-9.

[6] A. H. Nasution, A. Onan, Y. Murapenulis, W. Monika, dan A. Hanafiah, "Benchmarking Open-Source Large Language Models for Sentiment and Emotion Classification in Indonesian Tweets," IEEE Access, vol. 13, hal. 94009-94025, 2025, doi: 10.1109/ACCESS.2025.3574629.

[7] G. I. Winata et al., "NusaX: Multilingual Parallel Sentiment Dataset for 10 Indonesian Local Languages," dalam Proc. 17th Conf. European Chapter Assoc. Comput. Linguistics (EACL), 2023, hal. 815-834, doi: 10.18653/v1/2023.eacl-main.57.

[8] C. Shaw, P. LaCasse, dan L. Champagne, "Exploring emotion classification of indonesian tweets using large scale transfer learning via IndoBERT," Social Network Analysis and Mining, vol. 15, no. 1, Art. no. 22, 2025, doi: 10.1007/s13278-025-01439-6.

[9] C. Smith et al., "Social media perceptions of college football performance and season length," PLOS ONE, vol. 20, no. 7, Art. no. e0325840, 2025, doi: 10.1371/journal.pone.0325840.

[10] J. Walker et al., "Psychology at the screen: Investigating the current VAR protocol," PLOS ONE, vol. 21, no. 4, Art. no. e0345704, 2026, doi: 10.1371/journal.pone.0345704.

[11] M. N. Nazarudin et al., "VAR consultation patterns and their association with fouls and misconduct: An analysis of the top five European football leagues," PLOS ONE, vol. 20, no. 10, Art. no. e0334518, 2025, doi: 10.1371/journal.pone.0334518.

[12] D. Suhartono, W. Wongso, dan A. T. Handoyo, "IdSarcasm: Benchmarking and Evaluating Language Models for Indonesian Sarcasm Detection," IEEE Access, vol. 12, hal. 87323-87332, 2024, doi: 10.1109/ACCESS.2024.3416955.

[13] M. A. Rosid, D. Siahaan, dan A. Saikhu, "Sarcasm Detection in Indonesian-English Code-Mixed Text Using Multihead Attention-Based Convolutional and Bi-Directional GRU," IEEE Access, vol. 12, hal. 137063-137079, 2024, doi: 10.1109/ACCESS.2024.3436107.

[14] M. Mujahid et al., "Data oversampling and imbalanced datasets: an investigation of performance for machine learning and feature engineering," Journal of Big Data, vol. 11, Art. no. 87, 2024, doi: 10.1186/s40537-024-00943-4.

[15] M. Bal, A. G. Kara Aydemir, dan M. Coskun, "Exploring YouTube content creators' perspectives on generative AI in language learning: Insights through opinion mining and sentiment analysis," PLOS ONE, vol. 19, no. 9, Art. no. e0308096, 2024, doi: 10.1371/journal.pone.0308096.

[16] K. Aziz et al., "Unifying aspect-based sentiment analysis BERT and multi-layered graph convolutional networks for comprehensive sentiment dissection," Scientific Reports, vol. 14, no. 1, Art. no. 14646, 2024, doi: 10.1038/s41598-024-61886-7.

Published

16-11-2026

How to Cite

Yasin, S. A., Josua Eriek Kurniawan, Raditya Fernanda Putra, Yumi Novita Dewi, Waeisul Bismi, & Suharyanto. (2026). ANALISIS SENTIMEN KOMENTAR YOUTUBE PERTANDINGAN PORTUGAL MELAWAN KROASIA MENGGUNAKAN TRANSFER LEARNING INDOBERT. Jurnal Ilmiah Informatika, 11(2), 230–241. https://doi.org/10.35316/jimi.v11i2.10816