Mapping Indian Social Science Research to SDGs: An Automated Machine Learning Framework

This research study reports progress on a minor research project funded by the ICSSR, aimed at categorising Indian social science research publications according to the United Nations’ 17 Sustainable Development Goals (SDGs) using Machine Learning (ML)- based categorisation. To build the system, a dataset of 200,000 bibliographic records (during the period from 2016 to 2024) was assembled initially from major databases, including OpenAlex, Lens, Dimensions (open access), and Scopus, Web of Science (commercial), by applying an array of filters to check that at least one author is from India. These bibliographic records underwent manual classification based on titles and abstracts, with certainty/confidence scores assigned to ensure data reliability by a team of research scholars in LIS. Subsequently, 100,000 quality records (with confidence score ≥0.50) were utilised to train selected ML backends (mainly associative models) available in the Annif open source framework. Performance evaluation revealed that a Neural Network backend, tuned through the hyperparameter optimisation utility of Annif, outperformed other ML backends in both metrics, like F1@5 and NDCG. The study further applied this system to map recent research trends (social science publications from India during January to June, 2025), revealing that Clean Water and Sanitation (Goal 6) emerged as the highest focused theme. This was closely connected to ‘People’ centric goals such as Good Health, Quality Education, and Gender Equality. It reveals that Indian social science scholarship prioritises immediate human welfare and essential infrastructure over environmental goals such as Climate Action. These findings demonstrate the feasibility of this automated prototype for assessing the nation’s contribution to these global development targets.