Mapping Public Opinion Topics on Educational Policy on Twitter Based on Geospatial Aspects Using Latent Dirichlet Allocation (LDA)

Educational policy has emerged as a prominent public issue that is widely discussed across social media platforms. As an open-access microblogging platform, Twitter generates a substantial volume of user-generated content that reflects public opinion in real time. Such data provide valuable insights for understanding societal responses to government policies. This study aims to examine the topic segmentation of public opinion regarding educational policies in Indonesia by employing the Latent Dirichlet Allocation (LDA) model. The dataset comprised 8,030 Indonesian-language tweets collected using education-related keywords. After a relevance filtering process, 699 tweets were retained for analysis. The text preprocessing procedures included case folding, removal of numerical characters and punctuation marks, elimination of Indonesian stopwords, and stemming to normalize word forms. The cleaned corpus was then transformed into a Document–Term Matrix (DTM) representation prior to topic modeling. LDA was applied with three predefined topics to extract latent thematic structures within the dataset. The results reveal that public discourse can be categorized into three principal themes: (1) government policies and national conditions, (2) government performance and public policy implementation, and (3) higher education issues related to university students. The topic distribution indicates that discussions concerning government policy and higher education issues are the most dominant themes within public conversations on Twitter. These findings contribute to the growing body of research on social media analytics in public policy studies and provide empirical evidence that may assist policymakers in identifying public concerns and evaluating policy communication strategies