Research On Personalized Exercise Recommendation Methods Based On The Graph Neural Network

Personalized exercise question recommendation is one of the hot research directions in the field of smart education.At present,the application of cognitive diagnosis method and deep learning method to personalized exercise item recommendation can achieve better recommendation effect.However,in the past work,when modeling students' ability level in the process of cognitive diagnosis,the high-level interactive information of students' exercise questions was ignored,resulting in insufficient modeling accuracy and insufficient interpretability of the recommended results.Cognitive diagnosis and prediction of students' problem-taking are key steps in personalized exercise-item recommendation.In the process of exercise question recommendation,there will be problems that the accuracy of modeling students' abilities is insufficient and the recommendation results are difficult to ensure interpretability and accuracy at the same time.This thesis analyzes the current research status and deficiencies in the field of smart education at home and abroad,and proposes a new cognitive diagnosis model and a personalized exercise item recommendation model.The main work of this thesis includes the following aspects:1.Aiming at the problem of insufficient model accuracy caused by ignoring the highorder interactive information of students' exercise questions in the process of cognitive diagnosis,a cognitive diagnostic model GCCD based on graph convolutional network is proposed.Connectivity,which aggregates the interactive features of students' exercise questions in the stage of building a student ability model,improves the expressive ability of the model,and adopts the monotonicity assumption and Q matrix to ensure the interpretability of cognitive diagnosis results.This model is exerciseed on the usual exercise data set of college students' data structure course,and the accuracy rate reaches 0.937,and the RMSE drops to0.222,which proves the effectiveness of the model proposed in this thesis.At the same time,this model also verified the effect of different layers on the performance of the model through graph convolution layer experiments,and conducted consistency experiments to verify that the interpretability of the model was better than other comparison models.2.A personalized test question recommendation model based on knowledge tracking was proposed in this thesis,constructs a student ability model to ensure its interpretability by using GGCD cognitive diagnosis model,and predicts the student's answer by tracking the change process of student ability with time series through Bi GRU,and uses the characteristics of students' historical question sequence to improve the prediction accuracy of the model.Experiments were conducted on the public datasets ASSISTments2009 and ASSISTments2017,and the AUC values of this model reached 0.903 and 0.931,respectively.At the same time,the consistency experiment on the two datasets verifies the solvability effect brought by the cognitive diagnosis model,and the validity of the difficulty setting of the exercise questions of this model is verified through the difficulty range parameter experiment.3.Personalized exercise question recommendation system was design and implement,which integrates system functions such as student answer questions,test question recommendation,and visual display of student ability.The system is built with Spring Cloud microservice technology,and the recommendation results generated by the personalized exercise question recommendation model are processed through the Hbase database.