Using NLP Methods in Social Sciences - Experience and Opportunities Presented through the Lens of the ONTOLISST Project

In the Ontolisst project, our contribution was threefold. First, during data processing and the creation of the LiSST thesaurus, we applied clustering and topic modeling to uncover patterns in large datasets, using generative AI for cluster labeling. Second, we validated LiSST by processing extensive sets of social science paper keywords and applying semi-supervised clustering for codebook validation. Finally, we developed automated labeling by fine-tuning XML-RoBERTa models to classify new survey questions and variables into the established codebook. In our presentation, we will briefly introduce these techniques and share key results, highlighting both our practical experience and the opportunities they open for social science research.