Mapping Four Decades of Postgraduate Physiology Research in Türkiye Using BERTopic-Based Neural Topic Modelling and Temporal Trend Analysis (1986–2025)

Background/Objectives: The aim of this study is to map the thematic structure of physiology-related postgraduate theses completed in Türkiye between 1986 and 2025 using a BERTopic-based neural topic modelling approach, and to examine the long-term temporal changes in research topics. Methods: A total of 2297 theses obtained from the Council of Higher Education (YÖK) National Thesis Center were subjected to bibliometric evaluation, and 2263 theses with suitable Turkish abstracts formed the text mining corpus. Thesis abstracts were represented using multilingual sentence embeddings; a topic modelling workflow based on BERTopic, UMAP and HDBSCAN was applied. Model selection involved screening 42 parameter configurations with seed = 42, followed by stability assessment of the 21 eligible configurations across five random seeds. Results: The final model identified 33 topics; 1491 theses were assigned to a topic, while 772 theses (34.1%) were classified as outliers. In temporal analyses conducted over five-year periods, a relative increase in representation was observed in three of the 19 topics meeting the evaluation criteria, while a decrease was noted in five. The strongest relative increase was observed in the topic of experimental obesity, nutrition and metabolic regulation (OR = 1.80; 95% CI: 1.40–2.30; q < 0.001). While normalized Shannon entropy increased over time, a sensitivity analysis with equal sample sizes indicated that the observed increase in diversity was substantially influenced by the marked increase in thesis volume. Conclusions: These findings describe long-term changes in relative topic representation among theses assigned to topics within the selected physiology department categories of Institutes of Health Sciences in Türkiye.