Journal DOI 10.68956/journal.human-performance-futures

Creative Commons source reading · Data quality and algorithm evaluation

Artificial intelligence for university physical education: a data–knowledge synergy digital-intelligent sports platform

Jiacheng Feng; Meijie Yao; Xueying Feng; Chenrui Wang; Hongmiao Chen; Dong Li; Qimeng Niu

Originally published in Front Public Health · 2026-09-08

From the article

Selected source text. JournalOS removes markup and shortens excerpts where indicated. The full article provides the complete context.

Abstract

This study presents the design, construction, and preliminary evaluation of a university-oriented digital-intelligent sports platform integrating a large language model with retrieval-augmented generation (RAG), campus data services, and domain knowledge resources. The platform architecture comprises technical, knowledge, operational-data, and user-application components supporting teaching assistance, exercise guidance, competition organization, and health-related information services. A formative campus deployment involving 428 students and six instructors documented implementation feasibility and user-facing application scenarios. To provide a more rigorous empirical supplement, an independent six-week, non-randomized, intact-class quasi-experiment was conducted with 196 students in four physical-education classes. Two classes…

Secondary and class-sensitivity findings

All six BPNSFS subdimensions favored platform-supported instruction after Holm correction. Adjusted differences ranged from 0.601 to 0.629 for the three satisfaction dimensions and from −0.815 to −0.576 for the three frustration dimensions (Supplementary Table S5). These findings suggest a broad pattern across autonomy, competence, and relatedness rather than a single isolated subdimension. Changes were directionally consistent in both platform-supported classes. Need-satisfaction mean changes were 0.689 in E1 and 0.637 in E2, compared with 0.038 in C1 and 0.011 in C2. Need-frustration changes were −0.608 in E1 and −0.705 in E2, compared with −0.034 in C1 and −0.002 in C2. Leave-one-class-out estimates remained in the same direction (Supplementary Table S6). The adjusted between-group difference in self-reported physical activity was…

Research conclusions

This study develops a university-oriented digital-intelligent sports platform centered on LLM + RAG services, domain knowledge organization, operational data, and user modules. The technical framework and operational data-flow description clarify how data acquisition, storage, retrieval, reasoning, application delivery, and feedback interact. The 428-participant deployment provides formative implementation evidence, while the independent six-week classroom study indicates favorable changes in students’ proximal psychological experiences during platform-supported instruction. No statistically significant between-group difference was identified for self-reported physical activity. Together, the findings support the feasibility of integrating the platform into university physical education, while larger multi-class and…

Editorial reading guide

  • What data created and tested the model, and was evaluation independent?
  • What is the cost of a false positive, false negative or out-of-distribution decision?