Topic Modeling on Social Media and News Website Data Concerning The Electronic-Based Government System in Indonesia

Abstract

The rapid growth of social media and news websites has generated a massive amount of unstructured text data on the internet, including public opinions regarding the Electronic-Based Government System (SPBE) in Indonesia. To discover hidden information and topics within this textual dataset, an effective analysis technique such as topic modeling is required. This study aims to map the discussion topics on social media (Twitter, Tiktok) and news websites related to SPBE, and to analyze the implementation of topic modeling using the Latent Dirichlet Allocation (LDA) method. This research employs a computational approach by applying the LDA algorithm to a text corpus consisting of 3,327 datasets. Model quality evaluation was conducted through experiments across 12 modeling scenarios using the Coherence Score metric. The results indicate that Scenario 12 is the most optimal model, achieving the highest Coherence Score of -1.234, which generated 5 main topics. This peak performance was reached under the configuration of K=5 topics, 100 iterations, an Alpha value of 0.001, and a Beta value of 0.001. The top five words with the highest weights dominating the dataset are "aplikasi" (145 times), "spbe" (95 times), "setuju" (51 times), "sistem" (51 times), and "elektronik" (47 times). In conclusion, this study proves that the LDA method is capable of coherently extracting latent topic structures, revealing that public focus is not only centered on the technical form of the SPBE application, but also highlights the system evaluation process and its impact on specific sectors such as educators (teachers).