A Temporal Sequence Framework based on Self-Attention for Student Dropout Prediction in MOOCs.

Student dropout prediction is a essential task for online education platforms to reduce dropout rates. Existing methods fail to capture temporal dependencies in student learning sequences, leading to the potential omission of information from distant past. Therefore, we propose a Gated recurrent unit Network model based on Self-Attention mechanism (SAGNet) to enhance the ability to focus on key contextual information and capture temporal trend for predicting the probability of students at risk of potential dropout. Our model experiments on a real dataset, and the experimental results demonstrate the superiority of the proposed method in student learning sequence prediction. Additionally, we perform performance experiment on temporal trend to demonstrate the effectiveness of our method. Compared with the existing model, SAGNet has the best performance, which has reasonable guiding significance for online education platform to formulate the intervention measures.