Topic Modelling Analysis on Indonesian News Using BERT Topic Model

This study applies the BERTopic model to analyze an Indonesian news corpus collected in December 2015, identifying and interpreting the primary topics discussed. By comparing the BERTopic approach with the traditional Latent Dirichlet Allocation (LDA) method, this research highlights the advantages of BERTopic in generating coherent and interpretable topics, especially for short and unstructured text data. The study evaluates the performance of BERTopic using data from major Indonesian news platforms, such as kompas.com, republika.co.id, viva.co.id, and tribunnews.com. The results reveal critical themes and trends in Indonesian news, demonstrating the capability of BERTopic to capture nuanced insights into public discourse. Unlike LDA, BERTopic leverages BERT embeddings and advanced clustering techniques, significantly improving topic coherence and interpretability. This research contributes to the growing literature on topic modeling by showcasing BERTopic's effectiveness in a linguistically and culturally diverse context. The findings offer valuable implications for researchers, journalists, and policymakers, emphasizing the importance of selecting appropriate topic modeling techniques for dynamic and complex datasets.