Deep Learning Cognitive Diagnostic Model Based on State Machines and Knowledge Graphs

Cognitive diagnostic methods use models and algorithms to assess a student’s mastery of specific knowledge or concepts. The research on cognitive diagnostic methods has developed a wealth of research results from the 1980s to the present. Among them, deep learning-based cognitive diagnostic methods are currently the mainstream methods in this research field, e.g., NeuralCD methods. However, since this type of method uses one-dimensional vectors to express knowledge points, it cannot reflect the complex interrelationships (e.g., forward and backward references, derived relationships) among knowledge points. Moreover, the deep learning process established by this type of method ignores the constraints of the structure of knowledge points on the solution results, so the final evaluation results of this type of method have a large deviation. As a result, this paper proposes a deep learning cognitive diagnosis model based on state machines and knowledge graphs(KGSCD). The model models the process of problem solving as a state machine, the steps of problem solving are regarded as the state change of the state machine, and the knowledge points required for the problem solving steps and the requirements of problem solving ability are taken as the conditions for the state change. In this paper, we use the Long Short-Term Memory (LSTM) model, use the state machine transfer matrix modified by the knowledge graph as the input, and finally realize the diagnosis of the assessor’s own ability and mastery of each knowledge point according to the similarity of the assessment results by comparing the assessed person with the ideal learner mastering the standard knowledge structure. This paper shows experimentally that KGSCD has higher prediction accuracy compared with similar methods.