Student engagement assessment using multimodal deep learning

Student engagement assessment plays an important role in enhancing students’ positive performance and optimizing teaching methods. In this paper, a multimodal deep learning framework is proposed for student engagement assessment. Based on this framework, we propose a method for engagement assessment that utilizes data from three modalities: video, text, and logs. This method implements the extraction of engagement indicators, the fusion of asynchronous data, the use of deep learning models to evaluate engagement levels, and the use of gradient magnitude mapping to further distinguish subtle differences between engagement levels. In subsequent empirical studies, we explore the applicability of several popular deep CNN models in this method and validate the reliability of the engagement quantification results using statistical methods. The analysis results demonstrate that the framework, which combines multimodal asynchronous data fusion and deep learning models to assess engagement, is both effective and practical.