Topic Modeling and Trend Analysis in Medical Education Literature Using Projective Non-Negative Matrix Factorization

In this study, we performed Non-Negative Matrix Factorization (NMF) based topic analysis on PubMed article abstracts in the field of medical education to analyze topic trends over the last 15 years. We also applied automatic frequency-based n-gram (contiguous word sequence) generation to enhance the results. The analysis begins with a pre-processing pipeline that performs Unicode normalization, text cleaning, stop-word removal and n-gram extraction. Projective Non-Negative Matrix Factorization (PNMF), a variant of NMF suitable for sparse data (i.e., data with many zero values), is then used to extract topics with relevant words and documents. Topic names were generated using Claude Artificial Intelligence (AI) to enhance interpretability. The analysis includes visual outputs showing topic distribution, yearly trends, and t-distributed Stochastic Neighbor Embedding (t-SNE) visualization. The number of topics was set to 10 manually to capture key themes from a dataset of 46,292 abstracts. The results reveal key topics in medical education, including the recent emergence of Generative AI and Large Language Models (LLMs) in the literature and an increase in equity-related studies over recent years.