Reviewing 55 years of Literature on Physics Education using Natural Language Processing

Literature reviews play a crucial role in informing the Physics Education Research (PER) community about the evolution of teaching practices and educational interests over time. Large-scale data analysis techniques offer novel possibilities for conducting comprehensive literature reviews. In this work, we applied an unsupervised natural language processing (NLP) method, called Latent Dirichlet Allocation (LDA), to a large dataset of articles from The Physics Teacher journal. Our goal was to identify and track thematic trends spanning nearly six decades (1963–2020). We found that the journal’s content has balanced discipline-specific topics, pedagogy and laboratory methods, and learning-theory perspectives, while placing increasing emphasis on students’ learning over time. We discuss the strengths of this approach and its limitations, including the need for contextual expert knowledge to interpret topics. Finally, we propose future directions for large-scale literature analysis in physics education.