ACD: Attention Driven Cognitive Diagnosis for New Learners Joining ITS

Bingdi Shao, Keai Wei, Longjiang Guo, Meirui Ren, Lichen Zhang, Peng Li

Abstract: In Intelligent Tutor Systems (ITS), Cognitive Diagnosis (CD) is an important and fundamental problem, which aims to discover learners’ proficiency in different knowledge concepts. However, existing CD models (CDMs) that are from the perspective of learners and scores ignore the cold start problem of new learners joining ITS (CSP for short). This paper proposes an attention mechanism-driven cognitive diagnosis model named ACD for new learners joining ITS to solve the cold start problem, which is composed of a three-layer attention mechanism neural network. Specifically, in the first layer, the cognitive state of part of the concepts was obtained using the attention mechanism on the learners’ exercises, concepts, and scores. In the second layer, attention is computed on all concepts and on the part of the cognitive state on concepts output in the first layer to obtain the cognitive state on all concepts. In the third layer, concepts, exercises, and the cognitive state of the learner output in the second layer was obtained using the attention mechanism on the learners’ scores on the exercises. Finally, the large number of experimental results on five real datasets show that ACD performs well on different evaluation metrics when new learners come into ACD.