Research on the analysis platform of the interactive mode of the collaborative learning group combining spatio-temporal graph convolutional network and YOLO real-time detection

This paper proposes a collaborative learning interactive mode analysis platform that integrates the spatio-temporal graph convolutional network and the YOLO real-time detection framework. Based on the SECI knowledge conversion theory, a spatio-temporal graph of student interaction is constructed to quantitatively analyze the dynamic characteristics in group discussions. Experiments show that the recognition accuracy of interaction modes of this model is 23% higher than that of manual coding and offline video annotation methods. The STGCN module realizes cross-scale analysis from micro-behaviors to macro-interaction modes through multi-scale spatio-temporal feature fusion. The YOLOv8 network structure module introduces the spatio-temporal graph convolutional attention mechanism and the RFCBAMConv lightweight module, which improves the accuracy and efficiency of target detection. In real classroom scenarios, the detection accuracy of the collaborative learning platform reaches 94.5%, the recall rate is 91.2%, and the F1 score is 0.928, which is 2.4–5.3% points higher than that of a single model. Through the collaboration between the spatio-temporal prediction module and YOLOv8, the model achieves the dual advantages of “high accuracy-low latency”, and the processing speed meets the requirements of real-time detection. Under the strong collaboration mode, the detection speed of the model is increased to 40FPS, the accuracy reaches 50.3%, the false detection rate is reduced to 8.7%, and the processing latency is reduced to 75ms. This study provides an extensible and interpretable analysis tool for collaborative learning research in the context of educational informatization.