Time-Aware Attentional Knowledge Tracing based on Pre-training and Feature Extraction
Yuxin Tian, Zhanquan Wang · 2023
Knowledge tracing (KT) has attracted increasing attention as the level of education informatization has increased. KT models students’ changing knowledge states over time based on their historical question responding and then forecasts students’ success in question answering. Many knowledge tracing models have been proposed to support the smart education system, but these models frequently fail to provide good interpretability, are insufficiently extracted for question information and frequently ignore the influence of the time factor on prediction. To address these concerns, we propose the Time-Aware Attentional Knowledge Tracing based on Pretraining and Feature Extraction (PFTKT), which is implemented in three steps: First, pre-train the question inputs to enrich the question representation. Second, extract student attributes to guide prediction. Finally, add time distance parameters to the attention mechanism to model students’ forgetting behavior. We conduct comprehensive tests on three real-world datasets to validate the model’s effectiveness, and the results reveal that PFTKT surpasses previous knowledge tracing models in terms of AUC scores, and we also validate the effectiveness of each important component of PFTKT.