AI-Driven Smart Transformation in Physical Education: Current Trends and Future Research Directions

Abstract

Although the rapid development of Artificial Intelligence (AI) in recent years has brought increasing academic attention to the intelligent transformation of physical education, the core knowledge structure of this field, such as its primary research topics, has yet to be systematically explored. The LDA (latent Dirichlet allocation) topic model can identify latent themes in large-scale textual data, helping researchers extract key research directions and development trends from extensive literature. This study is based on data from the Web of Science Core Collection and employs a systematic literature screening process, utilizing the LDA topic model for in-depth analysis of relevant literature to reveal the current status and trends of AI technology in physical education. The findings indicate that AI applications in this field primarily focus on three areas: “AI and data-driven optimization of physical education and training”, “computer vision and AI-based movement behavior recognition and training optimization”, and “AI and virtual technology-driven innovation and assessment in physical education”. An in-depth analysis of existing research shows that the intelligentization of physical education, particularly in school and athletic training contexts, not only promotes sustainable development in the field but also significantly enhances teaching quality and safety, allowing educators to utilize data more precisely to optimize teaching strategies. However, current research remains relatively broad and lacks more precise and robust data support. Therefore, this study critically examines the limitations of current research in the field and proposes key research directions for further advancing the intelligent transformation of physical education, providing a solid theoretical framework and guidance for future research.

1. Introduction

With the continuous advancement of technology, artificial intelligence (AI) has found increasingly deep applications across various fields, particularly in education, where it demonstrates tremendous potential. As a tool capable of optimizing resource allocation, improving teaching quality, and promoting personalized learning, AI equips educators with new resources and offers students a more efficient learning experience. In the field of physical education, traditional teaching methods often overlook individual differences among students, resulting in standardized training methods with limited personalized feedback [1]. To address this issue, it is essential to recognize the importance of a pedagogical approach that prioritizes the holistic development of students. Physical education should not merely focus on performance metrics but also foster cognitive, personal, and social skills, ensuring that all aspects of student growth are considered [2]. By designing personalized learning paths and conducting targeted training data analysis, AI can better address individual needs, enhancing the overall effectiveness of physical education. Through data-driven decision-making, AI provides teachers with precise insights to optimize their teaching strategies, allowing them to create more personalized plans based on each student’s abilities and performance, thereby improving the specificity and effectiveness of instruction [3].

With the rapid advancement of artificial intelligence technology, its application in the field of education, especially in physical education, is increasingly expanding and deepening [4]. This study focuses on the intersection of physical education and AI, using the LDA topic model to systematically identify the main research directions in AI applications within physical education. By revealing the primary themes and trends in current research, this study aims to provide educators and researchers with structured insights that promote the intelligent development of physical education.

The remainder of this study is structured as follows: Section 2 reviews relevant research on AI in the field of physical education, providing a theoretical foundation for subsequent analysis. Section 3 describes the process, methods, and results of determining the research themes. Section 4 discusses the current status of the research themes and summarizes various aspects. Section 5 outlines the contributions of this study and provides an outlook on the future development of AI integration in physical education.

2. Literature Review

2.1. Applications of AI in Physical Education

As an essential part of education, physical education not only aims to improve students’ physical fitness but also emphasizes their comprehensive development in cognitive, social, and psychological aspects [5]. Physical education is divided into four sub-disciplines: physical education training, sports humanities, exercise biology, and traditional ethnic sports [2]. Unlike sports training, which is oriented towards skill enhancement and competitive performance, physical education focuses on students’ holistic growth and lifelong participation in physical activities [6]. Therefore, the application of AI in physical education should be based on multidimensional educational goals to meet students’ diverse needs.

The application of Artificial Intelligence (AI) in physical education has become a significant research topic in the field of education. AI improves and innovates physical education methods through technical means to enhance teaching effectiveness and personalization [1]. It can improve educational quality through personalized feedback, optimization of teaching resources, and enhancing the overall learning experience, thereby compensating for the limitations of standardized teaching models in accommodating individual student differences [4]. The application of AI in personalized learning is particularly effective, as it provides targeted feedback and learning pathways to meet the needs of different students, thereby enhancing teaching effectiveness. The widespread use of wearable devices has deepened AI applications in physical education, as these devices can monitor students’ physical conditions, enhance teacher–student interactions, and develop personalized learning plans based on real-time data, ultimately improving overall teaching quality [7].

In the dynamic and unpredictable environment of physical education, the application of AI technology greatly alleviates spatial and temporal limitations, making intelligent and flexible physical education possible. For example, AI assists students in scientifically training during fragmented times and spaces, thereby enhancing training efficiency and effectiveness [1]. Additionally, the application of advanced technologies like Augmented Reality (AR) in physical education significantly improves educational quality, particularly in the acquisition and understanding of motor skills. By using 3D models to demonstrate complex movement details, students can understand learning steps from multiple perspectives [8]. The integration of AI with AR and Virtual Reality (VR) holds promise for providing more immersive and interactive experiences in physical education, enabling students to receive guidance within real movement contexts. Research shows that AR-assisted instruction has notable advantages in skill acquisition and motivation enhancement [9]. Through AR and VR technologies, students can simulate and practice physical skills in a virtual environment, receive real-time feedback, and experience personalized training, which increases their learning interest and improves their athletic skills [10].

The application of AI in physical education not only demonstrates unique advantages in improving teaching efficiency and personalization but also plays an active role in ensuring safety and teaching quality. In the future, the integration of AR and VR with AI is expected to continuously enhance the immersive and interactive aspects of physical education, supporting comprehensive and scientific physical training. Machine vision and 3D motion teaching positioning technologies will further improve the precision and effectiveness of athletic training, providing scientific support for physical education and training [11].

In recent years, there has been a growing academic interest in the field of intelligent physical education, resulting in a large body of related research. However, issues such as scattered research directions and content have also emerged in this field. Existing studies have attempted to analyze literature on the integration of Artificial Intelligence (AI) and physical education using traditional bibliometric methods. Although the study by Lee and Lee [12] offers valuable insights into this field, it primarily relies on quantitative statistical data and lacks in-depth thematic exploration of text content, potentially overlooking certain research themes. This approach has limitations in revealing the breadth and evolution of topics and may not fully reflect the application trends and complexities of AI in physical education. To address these limitations, this study employs the LDA topic model to systematically analyze the relevant literature, aiming to explore the primary research directions in the integration of AI and physical education through a broader data pool. This approach not only captures the thematic distribution within the existing literature but also better illustrates the current applications and future potential of AI in physical education, providing a more comprehensive theoretical framework for subsequent research.

2.2. Applications and Advantages of LDA Topic Modeling

In recent years, with the development of big data technologies and text mining tools, researchers have increasingly adopted data-driven methods to analyze large-scale text data, aiming to reveal key themes and development trends within specific research fields. This data-driven approach has been widely applied in academia, especially in the natural sciences, technical sciences, and health sciences, where methods like keyword network analysis are used to understand the structure and evolution of scientific knowledge. This approach not only reveals research topics across different disciplines but also helps researchers identify potential research hotspots and future trends, enhancing the understanding of disciplinary development [13,14]. However, keyword network analysis has certain limitations: many articles may lack keywords, or the keywords provided may be selected from a predefined list, making it difficult to accurately reflect the actual content and research focus of the articles [15]. With the rapid advancement of artificial intelligence in recent years, traditional literature analysis methods such as keyword networks have gradually struggled to handle high-dimensional data and address complex problems. Consequently, researchers have turned to natural language processing (NLP) methods to study scientific documents in greater depth, providing more comprehensive support for scientific research. Among these techniques, the LDA topic model, a prominent text mining tool, has garnered significant attention in academia [16,17]. The LDA topic model can identify latent topics within large-scale text data, aiding researchers in extracting key research directions and development trends from extensive literature [18]. Compared to traditional qualitative literature reviews, the LDA model provides higher systematization and objectivity in a data-driven manner, capable of revealing hidden thematic structures.

The advantages of LDA topic modeling have been widely applied across various disciplines, especially in research that requires the analysis of large text datasets, such as sociology and education [19,20]. Hamed Jelodar et al. [17] have also demonstrated that the LDA model can effectively identify hidden topics within large-scale documents by analyzing word frequency and distribution patterns. It has been extensively applied in complex fields such as software engineering, political science, healthcare, and linguistics and is suitable for a variety of contexts, including social media and academic research.

In this study, we apply the LDA topic model to analyze research directions on AI in physical education, providing a more systematic perspective for research and practice in the field. Compared to traditional methods, the LDA model reveals latent thematic structures, helping to capture AI application trends and its future potential comprehensively. This study aims to provide theoretical support for the intelligent development of physical education and to encourage further research in this area.

3. Methods and Materials

3.1. Data Sources and Research Methods

This study aims to analyze the current achievements, future development directions, and potential challenges in the integration of “artificial intelligence and physical education”. We selected journal articles from the Social Sciences Citation Index (SSCI) and Science Citation Index (SCI) in the Web of Science (WOS) Core Collection as the data sources, reflecting a focus on rigor and credibility [21,22]. Our search query was constructed as TS = ((“Artificial intelligence” OR “Artificial neural network” OR “case-based reasoning” OR “cognitive computing” OR “cognitive science” OR “computer vision” OR “data mining” OR “data science” OR “deep learning” OR “expert system” OR “fuzzy linguistic modelling” OR “fuzzy logic” OR “genetic algorithm” OR “image recognition” OR “k-means” OR “knowledge-based system” OR “logic programming” OR “machine learning” OR “machine vision” OR “natural language processing” OR “neural network” OR “pattern recognition” OR “recommendation system” OR “recommender system” OR “semantic network” OR “speech recognition” OR “support vector machine” OR “SVM” OR “text mining”) AND (“physical education” OR “sports education” OR “fitness education” OR “sports training”)). Although the search strategy includes various interdisciplinary keywords, this study has rigorously screened the data to ensure that all selected literature is directly related to the application of artificial intelligence in physical education. The keywords “cognitive computing” and “communication codes” have practical applications in personalized feedback and learning analytics within physical education. By incorporating these keywords, we are able to comprehensively capture the diverse application scenarios and technological needs in the intelligent transformation of physical education [17]. This search covered the period from 1 January 2003 to 1 August 2024. We chose this timeframe because, since 2003, the field of AI integration with physical education has seen a significant increase in attention. This period allows us to observe the development of technology, policy support, and emerging trends and challenges in practice. The cutoff date for the search was 1 August 2024, when we conducted the final literature search and data collection. The search strategy was designed to cover all potentially relevant literature on the research topic. However, despite the broad scope, the search may include some articles that are less relevant to the topic. The framework of Data sources and research methods is shown in Figure 1.

We implemented a multi-round filtering process to ensure the quality of the selected literature. The initial search resulted in 652 articles. In the first round of screening, the primary author reviewed the titles, abstracts, and relevance to the research topic, eliminating approximately 381 irrelevant or off-topic articles, reducing the pool to 271 articles. In the second round, three members of the research team systematically evaluated the remaining articles, focusing on thematic alignment and relevance to the research objectives, removing 124 articles that, despite appearing relevant, did not meet the content requirements of the study [23]. In the third round, a cross-review of the remaining 147 articles was conducted, with several experts reviewing to eliminate articles with high redundancy or insufficient connection to the core topic, leading to the exclusion of 51 more articles. After multiple rounds of rigorous filtering and review, a final set of 96 eligible journal articles was selected for analysis.

After constructing the dataset, we performed text preprocessing, including stemming and thematic summarization. Next, we applied the LDA topic model to identify key themes in the text and used the perplexity metric to determine the optimal number of topics in the model. By analyzing the probability distribution of topic words, we explored the research hotspot of “AI integration with physical education” through visualization analysis. The framework of the study is shown in Figure 2.

3.2. LDA Topic Model

Latent Dirichlet allocation (LDA) is a probabilistic Bayesian model for analyzing discrete data and is one of the most effective techniques in text mining. It is widely used for data mining, uncovering latent data, and identifying relationships between data and text documents [17]. The core feature of the traditional LDA model is its unsupervised nature, which enables it to autonomously execute multiple analysis steps with minimal human intervention, even achieving automatic labeling functionality [24].

The LDA topic model is one of the most effective techniques in text mining and has been applied in numerous disciplines and industries. For instance, significant results related to LDA have been achieved in fields such as biomedical research, communication studies, and maritime target detection [17,25].

In this study, we employed the LDA topic model based on Dirichlet distribution because of its superior performance in handling large volumes of documents and interpreting identified latent themes, compared to several other algorithms [26]. This approach effectively addresses the challenges posed by co-occurrence and keyword analysis in traditional bibliometric studies. In many cases, documents may not explicitly list keywords or the keywords are merely selected from a predefined list, which significantly limits their accuracy and affects the true and comprehensive representation of the document’s content [15].

Topic modeling infers latent themes from textual data and automatically identifies topics within articles, showing their distribution across different documents. This method does not rely on predefined keywords or co-citation relationships but instead extracts information directly from the text, enabling a more accurate capture of the document’s meaning [27].

Figure 3 illustrates the generative process of the latent Dirichlet allocation (LDA) model [27], which assumes that each document is a mixture of several topics and that each topic is a distribution over multiple words. In this model, α and β control two Dirichlet distributions. The parameter α randomly generates the topic multinomial distribution θ for each document, θ then randomly generates a topic z, and β randomly generates the word multinomial distribution φ for the corresponding topic. Combining the topic z and the corresponding word distribution φ generates a word w. This process continues until a document with m words is generated, ultimately leading to n documents under k topics [28]. Through this model, we obtain the document-topic and topic-word distributions related to the integration of artificial intelligence and physical education.

The document–topic distribution helps analyze the directions and focus areas of AI integration with physical education. By comparing the topic distributions θm of different documents, we can identify which topics are frequently associated with integrating AI and physical education. Analyzing the document–topic distribution also helps uncover new research directions and unexplored area [17]. Researchers can determine new research avenues and propose recommendations by identifying topics that appear less frequently in existing documents.

Similarly, the topic–word distribution provides a deeper understanding of the research content in the AI and physical education domain. By comparing the word distributions within different topics, we can identify φk and recognize which words are frequently associated with AI and physical education integration [28]. This analysis not only helps determine future research directions but also provides new suggestions for policymakers, highlighting their crucial role in shaping the future of AI in physical education [17].

The above analysis highlights the core themes and discussion points of AI integration with physical education. This approach provides a deeper understanding of current research and guides future research directions and policy-making, ultimately promoting the realization of AI integration in physical education.

3.3. Paradoxical Leadership

The LDA algorithm begins with setting parameters, including the prior parameters α and β from the Dirichlet distribution, as well as the number of topics, k. In this study, the values of α and β were set to 0.1 and 0.01, respectively, which are common settings in the literature [29].

Since the number of topics k significantly affects the estimated topics, selecting the appropriate number of topics is crucial. We used topic perplexity to determine the optimal number of topics.

(1) Topic Perplexity: When the cosine similarity between topics decreases as the number of topics increases, there may be an issue of over-clustering. Therefore, we introduced perplexity as a measure to reduce such problems. Perplexity is a standard method for measuring the predictive power of the LDA model [30]. It is expressed by the following Formula (1).

$$Perplexity=\mathit{exp}\left\{-{\displaystyle \frac{{\sum }_{d=1}^M{\mathit{log}}_P\left(w_d\right)}{{\sum }_{d=1}^m{N}_d}}\right\}$$

(1)

The LDA topic model requires the number of topics to be predefined. A widely recognized method for determining this is by using perplexity, a key indicator for selecting the optimal number of topics. As illustrated in Figure 4, the model achieves its lowest perplexity when the number of topics is set to 3. This finding is further supported by the pyLDAvis visualization results (pyLDAvis version 3.3.1) presented on the left side of Figure 5, we observe that when the number of topics is set to 3, there is minimal overlap between topics, indicating better classification performance. Therefore, the number of topics in this study was finalized as 3. The right side of Figure 5 shows the top 30 words most closely associated with Topic 1. By entering different queries in the “Selected Topic” text box in the upper left corner of Figure 5, the top 30 words with the highest correlation for each topic can be displayed.

4. Data Results and Analysis

4.1. Data Results

After completing the LDA model training, we obtained two important output files: the “document-topic” distribution and the “topic-word” distribution. To identify and label each topic effectively, we first analyzed the top-ranking words under each topic in the “topic-word” distribution. We prioritized the top 15 high-probability words in each topic and used these keywords to determine the core content of the topic [19]. However, certain words, despite appearing at the top, may not truly represent the core content of the topic. For example, common or stop words (such as “the”, “is”, “in”, etc.) may appear in high-probability positions but have weak distinguishing power for the topic. In such cases, considering words further down the list of high-probability terms can provide more detail and insight into the topic, helping to accurately interpret and label it. By selecting the top 15 high-probability words, we can capture the core essence of the topic while maintaining flexibility in addressing the influence of stop words.

During this process, we excluded vague or repetitive topics to ensure each identified topic is unique and clearly defined. By organizing and analyzing these high-probability words, we assigned appropriate labels to each topic.

Table 1. Distribution of topics and high-probability feature words.

Number	Topic Identification Categories	Top 15 High-Probability Feature Words of the Topic
Topic1	AI and Data-Driven Optimization of Physical Education and Training	Network, student, college, mining, neural, design, deep, classroom, development, research, PE, analysis, BP, time, paper
Topic2	Sports behavior recognition and sports training optimization based on computer vision and AI	Recognition, computer, action, human, image, athlete, motion, vision, paper, feature, machine, network, accuracy, interaction, movement
Topic3	Innovation and Evaluation of Physical Education Driven by Artificial Intelligence and Virtual Technology	Student, AI, evaluation, analysis, information, improve, machine, virtual, quality, college, research, activity, PE, performance, teacher

Explanatory note: PE = physical education; BP = Back Propagation.

presents the three main topics identified through the LDA topic model related to integrating artificial intelligence and physical education and the distribution of their corresponding high-probability words. This provides a basis for further understanding each topic’s core content and characteristics.

For example, in Topic 1 from Table 1, high-probability words include “algorithm”, “network”, “system”, “BP”, “mining”, and “intelligent”. These terms suggest that this topic focuses on the development of neural network algorithms and data mining technologies in artificial intelligence, as well as their application in physical education. Therefore, we have labeled this topic as “AI and Data-Driven Optimization of Physical Education and Training”. In Topic 2, high-probability words such as “students”, “learning”, “AI”, “college”, “quality”, and “improve” indicate that the topic explores the role of artificial intelligence in enhancing the learning quality of college students, particularly in connection with physical education. Hence, this topic has been labeled “Motion Behavior Recognition and Sports Training Optimization Based on Computer Vision and AI”. Topic 3 includes high-probability words like “recognition”, “learning”, “deep”, “motion”, “system”, and “human”, highlighting the application of recognition systems and deep learning technologies in sports training. Based on these keywords, this topic has been named “AI and Virtual Technology-Driven Innovation and Evaluation in Physical Education”.

4.2. Analysis of Hot Topics in the Integration of Artificial Intelligence and Physical Education

4.2.1. AI and Data-Driven Optimization of Physical Education and Training

With the rapid development of artificial intelligence (AI), various industries are actively seeking ways to integrate AI technologies and the field of physical education is no exception [33]. Increasingly, research is focusing on the application of AI in physical education, making exploring how AI can enhance teaching and training outcomes a prominent research area [7]. The advancement of AI technologies, particularly the application of neural networks, deep learning, and data mining, has profoundly impacted physical education and training [34,35]. In the simulation training results, high-probability keywords such as “network”, “student”, “college”, and “mining” indicate that scholars in this research area tend to explore the integration of neural networks, deep learning, and data mining technologies within physical education and training. These studies go beyond a superficial understanding of the broad concept of AI, delving into the application and future development of data mining (“mining”, “system”) and deep learning (“deep”) technologies in physical education and training. This research plays a critical role in unlocking the potential of current physical education and enhancing its sustainability. Data mining (“mining”, “system”) and deep learning technologies (“deep”) provide better technical support for physical education, making them crucial for optimizing both teaching and training processes [33]. Hence, Topic 1 has been labeled “AI and Data-Driven Optimization of Physical Education and Training”.

In recent years, many scholars have explored how AI technology can optimize physical education classes, analyze training data, and improve teaching outcomes through the application of neural network technologies. The studies by Li et al. [36] and Liu et al. [37] indicate that the integration of network technology (“network”) with higher education physical education not only enhances teaching effectiveness, promotes the smartification of physical education, optimizes resource allocation and boosts student participation but also offers new perspectives and methods for research in the field of physical education. This integration not only helps improve the current state of physical education but also has a profound impact on its future development. Their study shows that AI can not only compensate for the limitations of traditional education but also enhance students’ overall competencies, allowing them to grasp key concepts in real-world settings. Wan [5] specifically highlighted the promising application of BP neural networks in sports training, demonstrating that BP neural networks can enhance students’ physical functions, enabling them to perform at their best while showcasing the multiple benefits of physical exercise. Yang et al. [1] predicted that the integration of AI technology will lead to revolutionary changes in future physical education. The application of AI holds great potential, as it can provide precise, data-driven support for curriculum design and training evaluation in physical education. These studies provide a solid foundation for the integration of AI with physical education and clearly highlight the significant value of AI in this field.

These studies summarize the multiple advantages of artificial intelligence and data-driven technologies in physical education, which not only have a positive impact on students and teachers but also promote the intelligent transformation of the entire education system, enhancing the efficiency and engagement of teaching. As algorithms continue to improve and the trend toward smartification accelerates, the education system can leverage a variety of technologies and methods to promote sustainable development. These studies provide a theoretical foundation for the intelligent transformation of physical education and offer guidance for future educational reforms, particularly in improving the efficiency and effectiveness of physical education [33].

4.2.2. Motion Behavior Recognition and Sports Training Optimization Based on Computer Vision and AI

Driven by globalization and technological innovation, advancements in computer vision and artificial intelligence have profoundly impacted sports training and education. By analyzing high-probability keywords such as “recognition”, “computer”, “action”, “human”, and “vision”, it is evident that this topic focuses on the significant influence of AI and motion recognition technologies on student education and athlete training, particularly in areas such as athlete motion recognition, sports analysis, and training process optimization. In recent years, combining motion recognition technologies with AI has become a research hotspot as researchers explore how these technologies can enhance effectiveness and accuracy in sports training and education. Therefore, Topic 2 has been labeled as “Motion Behavior Recognition and Sports Training Optimization Based on Computer Vision and AI”.

Scholars within this topic examine the contributions of motion behavior recognition, enhanced by computer vision and AI technologies, to the optimization of sports training. Regarding the application of computer vision technology in motion behavior recognition, Liu et al. [38] pointed out that compared to manual methods, AI-powered motion behavior recognition not only improves accuracy and objectivity but also provides real-time feedback, helping reduce subjective bias. This enables students to adjust their posture and performance in a timely manner, improving teaching efficiency and promoting personalized instruction to some extent. Research by Lin and Song [39] showed that athlete motion recognition can enhance individual athletic performance and improve students’ overall abilities. When combined with human–computer interaction technologies, real-time communication between athletes and training equipment can be achieved, significantly increasing the precision and efficiency of training feedback. Lin and Song [39] emphasized that machine learning and motion recognition can improve training outcomes and efficiency, but integrating motion recognition results with machine learning technologies remains a challenge. Lv et al. [40] focused on using deep learning algorithms to enhance the accuracy and efficiency of motion recognition in sports training. The study noted that as deep learning technology advances, motion recognition will be applied more broadly across various real-world scenarios. Future research may further explore more efficient network architectures and training strategies, as well as cross-scenario and cross-perspective motion recognition techniques. Yang et al. [1] demonstrated that human–computer interaction technologies, when introduced into sports education and training, can assess students’ learning attitudes and interests, thereby boosting their enthusiasm for learning.

This topic highlights how motion behavior recognition technology, based on computer vision and artificial intelligence, is driving revolutionary advancements in sports training and education. It improves the efficiency and accuracy of motion recognition and provides technical support for creating personalized training programs, demonstrating broad application prospects. Additionally, AI technology has significantly enhanced the fairness of physical education. Intelligent evaluation systems offer precise, unbiased feedback to each student, increasing transparency in teaching, reducing the impact of human bias, and ensuring the fair allocation of educational resources. This, in turn, promotes the comprehensive development of students, making physical education more equitable and inclusive [41,42].

4.2.3. AI and Virtual Technology-Driven Innovation and Evaluation in Physical Education

The development of artificial intelligence has brought profound changes to the field of physical education. In particular, the combination of generative AI and virtual reality (VR) technologies has not only transformed traditional teaching models but also revolutionized aspects such as teaching quality, efficiency, and the evaluation of student performance. Generative AI can generate personalized training plans based on individual student data, simulate various sports scenarios in real-time, and provide dynamic feedback to students and teachers, optimizing physical education programs. This technology helps teachers create innovative teaching methods tailored to individual student needs and enables more accurate evaluations. By analyzing high-probability keywords such as “student”, “AI”, “evaluation”, “analysis”, “information”, and “improve”, it is clear that this topic focuses on the impact of AI and VR technologies on physical education, particularly regarding the formulation of teaching plans and the innovation of teaching methods. Therefore, Topic 3 has been labeled as “AI and Virtual Technology-Driven Innovation and Evaluation in Physical Education”.

Regarding the application of artificial intelligence in physical education, Wang [35] pointed out that by introducing data mining technologies, universities can more effectively manage and analyze students’ physical performance data. Additionally, data visualization transforms complex data into simple and easily understandable charts, helping students better understand their own performance. The visualized information also provides comprehensive data analysis and decision support, aiding teachers in delivering more precise guidance. Zhou et al. [4] explored the application of generative AI in physical education, demonstrating how it can optimize and adapt to the current educational structure, streamline classroom processes, and provide strong technical support for teachers by generating various plans to enhance teaching strategies. Similarly, Lee and Lee [12] found that generative AI can provide educators with real-time classroom insights and offer different strategies to address the varying states of learners, effectively assisting educators in decision-making and optimizing educational plans, ultimately improving teaching efficiency and quality. This technology enables more accurate data analysis and feedback, helping teachers better understand and master skills. It improves training quality and boosts student engagement in sports training. Regarding information management in physical education, Deng et al. [43] emphasized that the development of information technology has had a revolutionary impact on schools. Technologies such as big data analysis, cloud computing, and the Internet of Things (IoT) present unprecedented opportunities for school sports. By leveraging these advanced technologies, educators can more accurately understand students’ physical conditions, exercise habits, and interests, allowing them to provide personalized physical education and training plans tailored to individual needs.

Their research explores the positive impact of integrating artificial intelligence and virtual reality technologies into physical education from different perspectives, revealing the promising development potential of this field. Furthermore, these technologies provide continuous technical support for the sustainable development of this area.

In the intelligent transformation of physical education, artificial intelligence and virtual reality technologies significantly enhance teaching quality. They provide a more personalized and effective learning experience in systematic physical education while also increasing student engagement in physical activities through innovative technological means [2]. Furthermore, these technologies offer technical support for teachers in implementing intelligent teaching. AI enhances the accuracy of monitoring and evaluating students’ learning processes and offers real-time performance monitoring, enabling teachers to provide personalized guidance and optimize teaching strategies. Virtual reality technology helps immerse students in the learning environment, offering an unprecedented learning experience. In virtual classrooms, students can engage in sports training as if they were physically present, significantly reducing the reliance on physical spaces and equipment. This innovation not only addresses practical challenges such as venue and equipment constraints but also reduces environmental pressure, aligning with the principles of sustainable development [7]. In the future, as AI and virtual reality technologies continue to evolve, they will play an even more significant role in physical education, driving the development of more innovative and more personalized sports education.

5. Research Conclusions and Future Prospects

5.1. Research Contributions

This study systematically identifies three main application directions of artificial intelligence in physical education through LDA topic analysis. This thematic classification not only provides a structured framework for the field but also offers a clear direction for future research. Secondly, the study delves into the groundbreaking contributions of AI technology in automated movement behavior recognition and personalized feedback, highlighting the important role of AI in enhancing the scientific nature of sports training and the accuracy of feedback. Thirdly, the study presents the practical application value of integrating AI with virtual reality technology in physical education classrooms, providing data support for improving student engagement and the precision of teaching assessments. Finally, the research proposes future research directions from the perspectives of sustainability and personalization, emphasizing the potential of intelligent physical education in resource utilization, educational equity, and social benefits. Through these contributions, this study provides theoretical support and practical guidance for academic exploration and practical development in intelligent physical education.

5.2. The Current State and Trends of Intelligent Physical Education

The current research on the integration of artificial intelligence (AI) and physical education primarily focuses on three key thematic dimensions: “AI and Data-Driven Optimization of Physical Education and Training”, “Motion Behavior Recognition and Sports Training Optimization Based on Computer Vision and AI”, and “AI and Virtual Technology-Driven Innovation and Evaluation in Physical Education”. These themes highlight AI’s broad application and diversity in physical education, covering the positive impact of computer vision, data analysis, virtual simulation technologies, and deep learning on sports education [11,44,45].

The research reveals that the application of AI in physical education involves multiple complex technological layers, raising the bar for the technological proficiency of sports educators. Through an in-depth analysis of these three major themes, it becomes evident that physical education is rapidly advancing toward a more intelligent, personalized, and data-driven future.

Moreover, integrating computer vision and AI has led to revolutionary breakthroughs in motion behavior recognition and sports training optimization. By leveraging machine learning and image processing technologies, sports training can overcome the limitations of traditional manual assessments, enabling automated motion recognition and feedback. This transformation improves training accuracy and enhances scientific rigor, injecting new vitality into sports training [34,35].

At the same time, integrating artificial intelligence and virtual technologies is setting a new trend in the innovation and evaluation of physical education. In sports classrooms, the innovative application of AI and virtual reality (VR) technologies has not only significantly enhanced student engagement and learning experiences but also provided teachers with more precise tools for teaching evaluation and management through data analysis platforms. This transformation is driving the deeper development of intelligent physical education, injecting new energy into the field [46].

The fusion of AI and virtual technologies is leading a new wave of innovation in physical education evaluation. The application of AI and VR technologies in sports classrooms not only greatly increases student participation and enriches their learning experiences but also equips teachers with more accurate tools for teaching evaluation and management through data analysis platforms. This shift is deepening the progress of intelligent physical education, bringing new vitality to the field [10].

Although existing research on the integration of artificial intelligence and physical education has demonstrated the diversity and complexity of AI applications in this field and yielded significant results [13], much of the research remains focused on theoretical models and case studies, with limited exploration of practical implementation and feasibility. In particular, there is still a lack of theoretical foundation and practical cases regarding the application of core AI technologies, especially in combination with other technologies [46]. Furthermore, many studies are confined to specific sports or aspects of sports management, which, while enhancing relevance, lack comprehensive applications across different sports or even disciplines. Additionally, most research focuses on short-term outcomes, with little attention paid to long-term sustainability.

In conclusion, AI technology has become a significant driving force in the advancement of intelligent physical education. From training optimization and motion recognition to the construction of virtual teaching platforms, AI applications are comprehensively pushing physical education from traditional models toward intelligent systems. This trend not only reflects the urgent demand for technology-driven innovation in the field of physical education but also offers new pathways and opportunities for the sustainable development of physical education.

5.3. Future Research Outlook

Physical education has its unique characteristics, with each sport requiring specific skills, rules, and underlying values, which place particular demands on educational methods and approaches. While technological advancements have driven the smartification of physical education, this evolution presents both opportunities and challenges. Future research can focus on addressing these challenges.

The personalized application of AI in physical education still holds significant potential for development [12]. Future studies should focus on developing more refined personalized teaching plans, utilizing deep learning technologies to optimize and tailor content based on multidimensional data, such as students’ physical fitness and athletic performance [34]. This approach aims to significantly enhance teaching quality and guide the evolution of physical education toward a more advanced level of intelligence and personalization.

The deeper integration of virtual reality (VR) and AI is a key research direction for the future. Although VR technology has been preliminarily applied in physical education, its close integration with AI remains to be further explored. A critical research question is how to combine virtual training environments with AI-driven data analysis to enable real-time adjustments to teaching content, offering students an immersive and personalized training experience.

The sustainability of smart physical education also warrants more attention. It is essential to explore how the widespread application of AI in physical education can promote long-term educational equity and efficiency while minimizing environmental resource consumption. The application prospects of AI technology in physical education are vast, extending beyond just improving training outcomes to enhancing students’ learning experiences and educational equity. Future research directions should focus on how to effectively integrate AI technology into the educational system to promote the holistic development of students.