Evolution of Research Topics in Solid Oxide Fuel Cells: Semantic Clustering Based on Language Models

The scientific literature on solid oxide fuel cells (SOFCs) is characterized by high fragmentation, multidisciplinarity, and rapid shifts in research priorities. Given the large number of scientific papers, methods for automated analysis of the research field aimed at identifying promising, breakthrough topics are in high demand. Traditional topic modeling methods cope poorly with this task, since they do not account for contextual relations and lose semantic density in large collections of papers. This paper presents a reproducible pipeline for topical classification applied to the titles and abstracts of SOFC-related papers retrieved from the open database OpenAlex. Based on contextual vectorization of texts and dimensionality reduction (UMAP), density-based and agglomerative clustering were tested, along with partitioning of the field by the structure of the reachability curve (OPTICS). The clusters of scientific papers obtained with OPTICS were semantically labeled using a local language model. A stable structure of 70 topical segments covering the entire field of papers was identified. The annual dynamics of the number of papers per topic and the citation dynamics are demonstrated. A shift in researchers' scientific interest from classical materials (Ni–YSZ, YSZ, ceria-based electrolytes) toward new materials and fuels (ammonia, perovskite cathodes, proton conductors, exsolution of nanoparticles, additive manufacturing) is shown. Draft preprint (version 1), not yet peer-reviewed. Supplementary materials: Appendix A (40 materials-science topics) and Appendix B (complete list of 70 SOFC research topics).