<![CDATA[The rapid growth of scientific output in institutional repositories has created significant challenges for the efficient retrieval of information, particularly when searches rely solely on unstructured metadata. Although topic modelling has been widely applied to large bodies of text, little attention has been given to Indonesian-language repositories and metadata-only datasets harvested through standardized protocols. This study aims to address this issue by using Latent Dirichlet Allocation (LDA) to analyze the research landscape of the Widyatama University Repository, based on titles and abstracts that were collected automatically via the OAI-PMH protocol. The proposed methodology integrates the following processes: automated metadata harvesting; Indonesian-language text preprocessing; probabilistic topic modelling; and quantitative evaluation using coherence metrics, complemented by qualitative interpretability analysis. The experimental results show that the optimal model was achieved with 12 topics, giving a Coherence Score of 0.5546 categorized as 'Good'. This demonstrates that meaningful thematic structures can be extracted even from limited textual metadata. The identified topics reflect the university's main research areas, such as Marketing Management (12.5%), Auditing (12.4%), and Human Resource Management (12.1%), as well as specific domains like Informatics (6.7%). To enhance practical usability, the model outputs were deployed in an interactive, Streamlit-based dashboard enabling dynamic exploration of topic relationships and temporal trends. This study contributes to repository analytics by demonstrating how topic modelling driven by metadata can transform institutional repositories into intelligent systems for discovering knowledge, supporting the navigation of research, landscape analysis and evidence-based decision-making for academic management.]]>
