Multi-Task Transformer with LSTM Model for Question Set Generation

Amidst the COVID-19 pandemic, Educational Technology (EdTech) has emerged as a vital option for remote learning. This transition has ushered in technical innovations while revealing challenges, including the widening educational gaps attributed to educators' struggles in assessing student comprehension in self-directed learning. The extensively researched field of knowledge tracing, which models a student's knowledge state promises personalized education for effective learning. Originally reliant on statistical analysis, this approach has evolved with technological advances and extensive datasets. This research introduces a multitask transformer with Long Short-Term Memory (LSTM) architecture capable of knowledge tracing and problem-solving time prediction. The model's adaptability facilitates task customization yielding superior performance in knowledge tracing. Our study explores problem-solving time prediction to motivate students and support educators in assessment design. Detailed implementation is provided in this paper.