Capability-Aware Knowledge Tracing for Learner's Knowledge Mastery Modeling

Yanzhao Wang, Qing Xie, Mengzi Tang, Yuhan Wang, Jingling Yuan, Yongjian Liu

Abstract: Knowledge Tracing (KT) aims to model the knowledge mastery of learners by predicting their performance in exercises based on past learning activity data. Recently, KT models employing deep neural networks have exhibited outstanding performance due to their excellent capabilities in representing learner features. However, they still face two major limitations: (1) Most of them cannot effectively track learner’s knowledge mastery progression in diverse scenarios. (2) They neglect an important factor in real-world scenarios that the learner’s capability often plays a crucial role in answering the exercises. Therefore, we propose CAKT for simultaneously modeling learner knowledge mastery and capabilities. Specifically, this approach first uses a knowledge evolution module with a gating mechanism in memory networks to simulate knowledge evolution based on prior states and interactions. A capability extraction module is then applied to model the learners’ capability development process. Finally, collaborative prediction is performed utilizing knowledge mastery and capabilities. Extensive experiments on four benchmark datasets demonstrate CAKT’s superiority. The code is available at https://github.com/WHUTwyz/cakt.

External IDs:dblp:conf/icic/WangXTWYL25