Implementation of Information Retrieval System and Latent Dirichlet Allocation as a Service for SDGs Mapping

Abstract

The assignment of Sustainable Development Goals (SDGs) categories to research and community service (PPM) data at IPB University was initially conducted manually, resulting in a large portion of data lacking SDG labels and causing inconsistencies. This study proposes an automated SDG labeling approach using an Information Retrieval System (IR System) integrated with the KMS PPM IPB. The main contribution lies in the development of an automation framework that combines Latent Dirichlet Allocation (LDA) for topic modeling and semantic similarity based on triplet loss to measure semantic relevance between PPM documents and SDG categories. Furthermore, the system is implemented as a REST API-based service, enabling seamless integration with the KMS environment and supporting deployment in a real-world system context with improved scalability and interoperability. Evaluation results show an F1-score of 16%, indicating limitations in balancing precision and recall. However, the average response time of 5.48 seconds demonstrates acceptable performance for an integrated system. These findings suggest that the proposed approach can serve as an initial step toward automated SDG labeling in large-scale PPM data management.