Analysis of knowledge structures in AI research within sports science using topic modeling: Focusing on SCI journals (2016–2026)

This study elucidates the chronological shifts in the knowledge structure of artificial intelligence research within sports science over the past decade. Analyzing 2737 SCI articles from the Web of Science (2016–2026), the research employed LDA-based topic modeling via Python 3.10.9 across three distinct phases: Phase 1 (2016–2019), Phase 2 (2020–2022), and Phase 3 (2023–2026). The results indicate that ‘training’ and ‘performance’ remained the dominant keywords throughout, underscoring AI's primary role in optimizing athletic output. Specifically, Phase 1 focused on high-intensity training and early machine learning for motion classification, while Phase 2 saw a surge in sophisticated action recognition driven by deep learning and wearable sensors. Phase 3 emerged with a focus on multi-dimensional performance analysis and machine learning-based injury risk prediction. Ultimately, the global knowledge structure is categorized into four domains: wearable-based action recognition, fatigue monitoring, injury risk analysis (e.g., concussions), and the development of rehabilitation systems. These findings demonstrate that AI in sports is evolving from simple data recording toward real-time monitoring and precision-based injury prevention systems.