LDA-Based Topic Modeling of YouTube Comments: Uncovering Learner Discourse in Indonesian JavaScript Programming Tutorials
Year: 2026
Type: conference-paper
Abstract: Opinion-based, unstructured information that can be gleaned from comments on educational videos is rich and underutilised on YouTube. This study uses the Latent Dirichlet Allocation (LDA) method to automatically determine the topics of the dominant discussion in an Indonesian language JavaScript programming tutorial based on comments. The comments underwent a 12-step pipeline, involving Indonesian stemming using Sastrawi, followed by the representation of the comments by Bag-of-Words and modeling the represented comments with LDA for three different topic counts, measured by C_v coherence. The best model (C_v = 0.7147) interpreted three topics: JavaScript logic and switch-case programming, coding exercises and variable manipulation, and learning appreciation and content engagement. Non-negative Matrix Factorization (NMF) was also tested as a baseline; although the baseline model achieved a higher coherence score, the topics it generated were less diverse, and lacked the affective appreciation theme found in the LDA topics, thus LDA was retained as the primary model. The findings demonstrate that LDA, when applied to informal social media data from Indonesia, can generate meaningful and coherent topic structure for informal Indonesian social media, providing educators and content creators with unsupervised knowledge of learners' actions.
Institution: Binus University
Language: English
Subfield: General Social Sciences
Field: Social Sciences
Domain: Social Sciences
SDG: Quality education
Open Access status: hybrid
OpenAlex ID: W7204754987