The Journal of China Universities of Posts and Telecommunications ›› 2023, Vol. 30 ›› Issue (1): 17-27.doi: 10.19682/j.cnki.1005-8885.2023.2002

• Artificial intelligence • Previous Articles     Next Articles

Deep knowledge tracking algorithm based on forgetting law

Guo Xiangbo, Wang Jian, Huang Mengjie, Wang Minghui, Yang Jian, Yu Yongtao   

  1. 1. School of Computer and Artificial Intelligence, Zhengzhou University, Zhengzhou 450001, China
    2. Yunnan Key Laboratory of Smart City in Cyberspace Security, Yuxi Normal University, Yuxi 653100, China
  • Received:2021-07-26 Revised:2022-04-01 Accepted:2023-02-13 Online:2023-02-28 Published:2023-02-28
  • Contact: Wang Jian, E-mail:
  • Supported by:
    This work was supported by the National Natural Science Foundation of China (61972133); Plan for “1125" Innovation Leading Talent of Zhengzhou City ( 2019 ); the Opening Foundation of Yunnan Key Laboratory of Smart City in Cyberspace Security ( 202105AG070010 ); Zhengzhou University Professors' Assisting Enterprises' Innovation-Driven Development Project (32213409).

Abstract: Knowledge tracking (KT) algorithm, which can model the cognitive level of learners, is a fundamental artificial intelligence approach to solve the personalized learning problem in the field of education. The recently presented separated self-attentive neural knowledge tracing (SAINT) algorithm has got a great improvement on predictingthe accuracy of students' answers in comparison with the present other methods. However there is still potential to enhance its performance for it fails to effectively utilize temporal features. In this paper, an optimization algorithm for SAINT based on Ebbinghaus' law of forgetting was proposed which took temporal features into account. The proposed algorithm used forgetting law-based data binning to discretize the time information sequences, so as to obtain the temporal featuresin accordance with people's forgetting pattern. Then the temporal features were used as input in the decoder of SAINT model to improve its accuracy. Ablation experiments and comparison experiments were performed on the EdNet dataset in order to verify the effectiveness of the proposed model. Seen in the experimental results,it achieved higher area under curve (AUC) values than the other present representative knowledge tracing algorithms. It demonstrates that temporal featuresare necessary for KT algorithms if it can be properly dealt with.

Key words: knowledge tracking, neural network, attention mechanism, personalized learning, Ebbinghaus' law of forgetting

CLC Number: