Multi-modal Data Fusion and Deep Mining Technology Innovation for Attention Analysis in Early Childhood Classrooms

This thesis is set against the backdrop of attention research in early childhood classrooms, focusing on the innovative application of data mining technologies in multi-modal information processing. Aiming at the limitations of traditional eye-tracking data analysis, this study proposes a deep learning framework that integrates spatio-temporal features and attention mechanisms. By constructing a predictive model for children's attention fluctuations, this framework enables dynamic monitoring and precise analysis of children's attention states in classrooms. Experimental results show that this method significantly improves feature extraction and prediction accuracy compared with traditional methods, providing an innovative technical solution for quantitative research in the field of early childhood education.