A method of knowledge tracing based on gating-controlled and attention mechanism
Yumin Dong, Yongjian Zhang, Li Fei-Fei, Tingting Zhu · Research Square · 2023
Abstract Knowledge tracing (KT) task is designed to track the changes in students ’knowledge status according to the historical learning interaction of students, and predict students’ future learning effects. Most of the existing KT models only judge their knowledge status based on the performance of students in learning interaction. They didn’t fully notice the impact of students’ learning and forgetful behavior in the state of knowledge. The difficulty of practice difficulty on knowledge tracing. We proposed a framework called the attention knowledge tracing based on the gating-controlled mechanism (referred to as GMAKT). In GMAKT, two gating-controlled mechanisms are designed to clearly simulate the forgetting and learning behavior in the learning process of students. Through the design gating-controlled mechanism, the difficulty of interactive records and knowledge concepts is effectively combined to track the dynamic changes of students’ knowledge of knowledge concepts. The results of a large number of experiments show that the proposed framework is better than the most advanced model in the KT task. In addition, ablation research shows that the forgetting and learning mechanism of design significantly helps improve GMAKT’s performance.