Research Of Knowledge Diagnosis Methods Based On Graph Aggregation And Attention

Cognitive diagnosis is a classic task in the field of educational psychology,which aims to predict students’ performance on unknown questions based on past answering records of students,and further diagnose students’ mastery of various knowledge points.Cognitive diagnosis can be divided into static cognitive diagnosis and dynamic cognitive diagnosis according to whether time series information is applied.The dynamic cognitive diagnosis problem is also known as the knowledge tracking problem.This dissertation studies the problems of static cognitive diagnosis and dynamic cognitive diagnosis respectively.For static cognitive diagnosis,this paper explores the challenge of how to add knowledge point aggregation in the process of cognitive diagnosis.According to the current research status,this problem has not been discussed by the predecessors.Subsequently,this paper proposes our model CDGK,a static cognitive diagnostic model based on artificial neural networks.CDGK can effectively solve the two challenges of today’s static cognitive diagnosis problems.It effectively applies each component in the learning system and fully mines the information between students,exercises,and knowledge points? at the same time,CDGK also considers Knowledge points are aggregated to achieve better model performance.Finally,this paper designed a series of experiments on two public data sets.From the experimental results,CDGK’s performance in the three indicators of AUC,RMSE and accuracy all exceeded the baseline model compared.This paper also designs ablation experiments and interpretability analysis experiments to illustrate the rationality and interpretability of CDGK.For the problem of dynamic cognitive diagnosis,this dissertation first analyzes the problem that the research results based on deep neural network cannot deal with the historical time series of students and practice questions on the problem of knowledge tracking.In order to solve this problem,this paper proposes the model DAKTN,which integrates dynamic historical time series into the cognitive diagnosis model based on deep neural network through the design of pooling layer.Subsequently,in order to better mine time history sequence information,this paper further designs the attention mechanism unit,so as to achieve the purpose of adaptively learning the corresponding representation vector according to the corresponding students and practice questions.Finally,this paper designed a series of experiments on three public data sets.From the experimental results,DAKTN’s performance in AUC,RMSE and accuracy all exceeded the baseline model compared.This paper also designs ablation experiments and sensitivity analysis experiments to illustrate the rationality of our model.