Investigating Indonesian Public Discourse in YouTube-Based Science Communication: A Corpus-Based Computational Linguistics Approach
Keywords:
Computational Linguistics, Digital Science Communication, IndoBERTAbstract
Digital platforms have expanded science communication by enabling audiences to interact with scientific content. However, computational resources for examining Indonesian-language audience discourse remain limited. This study presents the Indonesian Public Discourse Corpus (IPDC), constructed from 202,429 YouTube comments from 35 science education videos on the Kok Bisa? channel. The research employed a multi-stage natural language processing workflow involving corpus acquisition, manual annotation, transformer-based classification, and full-corpus inference. A dataset of 10,500 comments was independently annotated into eight discourse-act categories: Question, Opinion, Disagreement, Correction, Suggestion, Praise, Agreement, and Experience. The annotation protocol calibration achieved 93.00% agreement (Cohen's Kappa = 0.9122). The strongest model, IndoBERT Base, achieved a Macro F1 of 97.40%, a Weighted F1 of 97.72%, and an accuracy of 97.73% on unseen Kok Bisa? videos. Full-corpus inference showed Opinion was the dominant discourse act (69.20%), followed by Question (22.59%). In contrast, Experience, representing only 2.18% of comments, received the highest mean number of likes. The findings demonstrate the potential of transformer-based classification for large-scale analysis of audience participation in Indonesian digital science communication.
ABSTRAK
Platform digital telah memperluas komunikasi sains dengan memungkinkan audiens berinteraksi dengan konten ilmiah. Namun, sumber daya komputasional untuk mengkaji wacana audiens berbahasa Indonesia masih terbatas. Penelitian ini menyajikan Indonesian Public Discourse Corpus (IPDC), yang dibangun dari 202.429 komentar YouTube pada 35 video edukasi sains di kanal Kok Bisa?. Penelitian menggunakan alur kerja natural language processing yang mencakup akuisisi korpus, anotasi manual, klasifikasi berbasis transformer, dan inferensi pada seluruh korpus. Sebanyak 10.500 komentar dianotasi secara independen menjadi delapan kategori wacana: Question, Opinion, Disagreement, Correction, Suggestion, Praise, Agreement, dan Experience. Kalibrasi protokol anotasi menghasilkan tingkat kesepakatan 93,00% (Cohen's Kappa = 0,9122). Model terbaik, IndoBERT Base, mencapai Macro F1 97,40%, Weighted F1 97,72%, dan akurasi 97,73% pada data uji dari video Kok Bisa? yang belum pernah dilihat model. Inferensi seluruh korpus menunjukkan Opinion sebagai wacana paling dominan (69,20%), diikuti Question (22,59%). Sebaliknya, kategori Experience (2,18%) memperoleh rata-rata jumlah likes tertinggi. Temuan ini membuktikan potensi klasifikasi berbasis transformer untuk menganalisis partisipasi audiens berskala besar dalam komunikasi sains digital di Indonesia.
References
Aji, A. F., Winata, G. I., Koto, F., Cahyawijaya, S., Romadhony, A., Mahendra, R., Kurniawan, K., Moeljadi, D., Prasojo, R. E., Baldwin, T., Lau, J. H., & Ruder, S. (2022). One country, 700+ languages: NLP challenges for underrepresented languages and dialects in Indonesia. Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 7226–7249. https://doi.org/10.18653/v1/2022.acl-long.500
Austin, J. L. (1962). How to do things with words. Oxford University Press.
Barik, A. M., Mahendra, R., & Adriani, M. (2019). Normalization of Indonesian-English code-mixed Twitter data. Proceedings of the 5th Workshop on Noisy User-Generated Text (W-NUT 2019), 417–424. https://doi.org/10.18653/v1/D19-5554
Bucchi, M., & Trench, B. (Eds.). (2021). Routledge handbook of public communication of science and technology (3rd ed.).
Routledge.
Cohen, J. (1960). A coefficient of agreement for nominal scales. Educational and Psychological Measurement, 20(1), 37–46. https://doi.org/10.1177/001316446002000104
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2019). BERT: Pre-training of deep bidirectional transformers for language understanding. Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), 4171–4186. https://doi.org/10.18653/v1/N19-1423
Dubovi, I., & Tabak, I. (2020). An empirical analysis of knowledge co-construction in YouTube comments. Computers & Education, 156, Article 103939. https://doi.org/10.1016/j.compedu.2020.103939
Dubovi, I., & Tabak, I. (2021). Interactions between emotional and cognitive engagement with science on YouTube. Public Understanding of Science, 30(6), 759–776. https://doi.org/10.1177/0963662521990848
Fauzan, M. F. D. (2026). Kok Bisa? physics discourse analytics: An interactive research companion for the Indonesian Public Discourse Corpus (IPDC) [Interactive web application]. GitHub Pages. Kok Bisa? Physics Discourse Analytics
Hill, V. M., Grant, W. J., McMahon, M. L., & Singhal, I. S. (2022). How prominent science communicators on YouTube understand the impact of their work. Frontiers in Communication, 7, Article 1014477. https://doi.org/10.3389/fcomm.2022.1014477
Koto, F., Rahimi, A., Lau, J. H., & Baldwin, T. (2020). IndoLEM and IndoBERT: A benchmark dataset and pre-trained language model for Indonesian NLP. Proceedings of the 28th International Conference on Computational Linguistics, 757–770. https://doi.org/10.18653/v1/2020.coling-main.66
Searle, J. R. (1969). Speech acts: An essay in the philosophy of language. Cambridge University Press.
Vosoughi, S., & Roy, D. (2016). Tweet acts: A speech act classifier for Twitter. Proceedings of the International AAAI Conference on Web and Social Media, 10(1), 711–714. https://doi.org/10.1609/icwsm.v10i1.14821
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Proceedings of Castle

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.


