Research On Cognitive Diagnostic Methods Based On Graph Attention Networks And Forgetting Mechanism

Nowadays,the rapid development of artificial intelligence,big data,and "Internet +" has ushered in a new era.With the popularization and in-depth application of the Internet,online education has become one of the popular new forms of teaching.It provides students with many online courses,making the popularization and promotion of personalized education possible.To explore the universality and extensibility of personalized education,it has become one of the research hotspots to study the cognitive diagnosis model of students’ mastery of knowledge concepts and answer performance.Existing studies on cognitive diagnosis show that the neural network modeling of student-exercise-knowledge concept to diagnose cognitive level can help students and teachers analyze students’ mastery of knowledge concepts from different angles and provide personalized recommendations and guidance of educational resources.The existing cognitive diagnosis methods have their advantages.However,there are still some shortcomings:(1)The existing cognitive diagnosis methods only focus on the interaction between studentsexercises-knowledge concepts but ignore the implicit correlation between knowledge concepts and the semantic information implied in the knowledge concepts,thus affecting the accuracy of predicting students’ answer performance;(2)The existing cognitive diagnostic methods are based on monotone hypothesis and ignore the forgetting characteristics of students’ learning,thus affecting the accuracy and interpretability of the assessment of student’s cognitive level.This paper tries to improve the cognitive diagnosis method based on a neural network from three aspects: "the implied correlation of knowledge concepts," "the semantic characteristics of knowledge concepts," and "the forgetting characteristics of students," and tries to improve the two deficiencies of the current cognitive diagnosis method.The main research work of this paper is as follows:(1)Join GAT And Word2 vec for Improving Education Cognitive DiagnosisThe existing cognitive diagnosis methods both ignore the implicit correlation between knowledge concepts and the large amount of semantic information hidden in knowledge concepts in them.Therefore,this paper proposes a Join GAT And Word2 vec for Improving Education Cognitive Diagnosis(JGATW).JGATW method introduces the implicit correlation of knowledge concepts and semantic information in neural network modeling to improve the performance and interpretability of existing cognitive diagnosis.Based on students’ historical answer records,this method extracts the relationship graph features and semantic features of knowledge concepts respectively through the graph attention neural networks and word2 vec and then makes a preliminary diagnosis of students’ mastery of knowledge concepts by combining the graph features and semantic features of knowledge concepts,and enhances the features through the attention mechanism.Then predict the accuracy of students’ answers to diagnose their mastery of knowledge concepts.Many experiments were conducted on two open-source datasets to compare experiments with popular cognitive diagnostic models.The results show that considering the implicit correlation between knowledge concepts and semantic information to improve the neural network method is helpful in improving the accuracy of cognitive level prediction.(2)Cognitive Diagnosis Based on Memory Forgetting Neural NetworkThe existing cognitive diagnostic methods are based on the monotone hypothesis but ignore the effect of forgetting in learning.Therefore,this paper improves the cognitive diagnosis method based on neural networks and proposes a Cognitive Diagnosis Based on Memory Forgetting Neural Network(LMCD).In this paper,we construct a memoryforgetting neural network layer(MFNN)that fully considers the influence of forgetting characteristics and potential features of student-exercise-knowledge concepts on students’ knowledge mastery.Based on the historical answer records of students,the LMCD method extracts the features of forgetting through the forgetting units and memory units of the memory-forgetting neural network layer and enhances the features by using the self-attention mechanism.Furthermore,the generalized matrix factorization is used to excavate the potential features between the student-exercise-knowledge concept and then integrate them with the features of forgetting.Finally,students’ cognitive level is diagnosed by predicting their answer performance.Many experiments were conducted on two open-source datasets to compare experiments with popular cognitive diagnostic models.The results show that the cognitive diagnosis method combining forgetting features and student-exercise-knowledge concept potential features can improve the performance of diagnosing students’ cognitive level.