Background and purpose With the exponential growth in the volume of competitive sports research, traditional bibliometric methods based on keyword co-occurrence struggle to accurately capture the field's deep semantic structure and evolutionary dynamics. This study aims to employ natural language processing (NLP) techniques to reconstruct the knowledge structure and evolutionary characteristics of competitive sports research from a micro-semantic perspective. Methods Utilizing the BERTopic topic modeling algorithm combined with dynamic regression analysis, we performed unstructured text mining on 24,659 core abstracts published between 2010 and 2024. Data were retrieved from the Web of Science, PubMed, and Scopus databases. Results The study reveals a distinct “monocentric, multi-dimensional” hierarchical structure in competitive sports research.Topic 0 (“Sports Injury and Load Management”) represents the largest volume and exhibits continuous, significant growth (“Hot”). It constitutes the field's epistemological foundation, demonstrating robust vitality driven by technological iteration. Meanwhile, the explosive growth of Topic 1 (“Sports Mental Health”) marks a profound paradigm shift from singular “physiological optimization” to “holistic psychophysical well-being."In contrast, Topic 4 (“Mega-event Management”) shows statistically non-significant growth, indicating that this sub-field has reached a stage of theoretical saturation. Furthermore, the evolution of the discipline can be categorized into three stages: the Budding Period (2010–2014), the Rapid Development Period (2015–2019), and the Multi-dimensional Transition Period (2020–2024). Conclusion Competitive sports science is currently in a critical transition period from reductionism to a complex systems perspective. Future research frontiers are likely to focus on the deep integration of biomedical technologies and humanistic psychology. This study provides a novel methodological perspective for understanding the evolutionary laws of the discipline.
Ling Pan (2026) studied this question.