Scholarlink

<p>To address the shortcomings of simplistic tactical recognition methods in team ball games, which neglect the complex interactions between players, we combine graph attention networks to model ball games as a spatiotemporal graph structure, with the ball and players as nodes. Graph convolution is used to capture the connections between nodes. Our evaluation is conducted on a dataset covering 14 categories of football and basketball tactics. Experimental results show that our proposed Tactic Graph Net model significantly outperforms other models in tactical recognition accuracy, achieving 91% in football and 92.3% in basketball. This demonstrates a clear advantage over long short-term neural networks and graph convolutional structures. Furthermore, we analyze the effectiveness of dynamic graph multi-task learning methods through ablation studies. In addition, we consider model performance under different scenarios, analyzing league styles and daily matches. Our model exhibits high advantages in model quantization deployment and accuracy robustness.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A team ball game tactical recognition and result prediction model based on graph neural networks and attention mechanisms

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月20日

查看全文
  • Yuzhen Yan,
  • Zhenjun Liang,
  • Zebo Qiao

摘要

To address the shortcomings of simplistic tactical recognition methods in team ball games, which neglect the complex interactions between players, we combine graph attention networks to model ball games as a spatiotemporal graph structure, with the ball and players as nodes. Graph convolution is used to capture the connections between nodes. Our evaluation is conducted on a dataset covering 14 categories of football and basketball tactics. Experimental results show that our proposed Tactic Graph Net model significantly outperforms other models in tactical recognition accuracy, achieving 91% in football and 92.3% in basketball. This demonstrates a clear advantage over long short-term neural networks and graph convolutional structures. Furthermore, we analyze the effectiveness of dynamic graph multi-task learning methods through ablation studies. In addition, we consider model performance under different scenarios, analyzing league styles and daily matches. Our model exhibits high advantages in model quantization deployment and accuracy robustness.