ChebNet-Based Automated Evaluation Framework for Public Physical Education Teaching Modes in Colleges

Most existing methods for analyzing and evaluating public physical education (PE) teaching modes in colleges and universities rely on subjective evaluation, lack systematic and data-driven automated analysis methods, and make it difficult to capture the complex interactive relationship between student behavior performance and courses and activities. This paper uses ChebNet to analyze the relationship between teaching mode and learning effect automatically, and combines graph node features and edge weight information to provide explainable evaluation results. First, a multimodal dataset structure is designed to integrate the multimodal data of curriculum design, student sports performance, teacher–student interaction records and teacher feedback into a unified graph model. Then, the high-dimensional representation of nodes and edges is learned from the graph structure through graph convolution operations. Finally, the high-order feature aggregation process introduces an adaptive weighting mechanism. By adaptively adjusting the convolution weights of different orders, the practical information in the high-order adjacency relationship is enhanced, revealing the complex relationship between students and teaching activities. The experimental results show that when the data ratio is 100%, the precision, recall and F-value results of ChebNet are 0.8783, 0.8701 and 0.8742, respectively, which are 5.88%, 5.94% and 5.93% higher than the GCN (Graph Convolutional Network) baseline model with better performance. The conclusion shows that ChebNet can effectively capture complex student behavior patterns, which helps to provide new perspectives and support for personalized PE and model optimization.