Modeling Political Discourse in Indonesia’s 2024 Election Using Unsupervised Machine Learning

The 2024 General Election in Indonesia has generated a large volume of diverse and unstructured digital political discourse, necessitating a machine learning-based analytical approach for efficient, objective, and scalable data processing. This study aims to map political discourse from 14,813 text data collected from the open-source "Indonesian Election 2024" dataset on the Hugging Face platform, encompassing social media posts (e.g., Twitter) and online news content from January to March 2024. This research integrates three core methods: Principal Component Analysis (PCA) for dimensionality reduction, K-Means for clustering, and Latent Dirichlet Allocation (LDA) for topic extraction. This combination represents an original approach in Indonesian political discourse studies, leveraging unsupervised learning techniques to enhance topic mapping efficiency compared to single-method approaches in prior research. The analysis identified three primary clusters electoral technical issues, candidate figures, and official agendas yielding a Silhouette Score of 0.51 (a clustering quality metric) and a top topic coherence score of 0.51. Validation was conducted both quantitatively and qualitatively by content experts. This approach not only demonstrates strong analytical capability in uncovering thematic patterns but also offers practical applications for institutions such as the General Elections Commission (KPU), Election Supervisory Body (Bawaslu), and the media in monitoring strategic issues and detecting potential disinformation in the lead-up to the election.