Research On Knowledge Tracing Algorithm Based On Dynamic Key-Value Memory Network

Knowledge tracing is one of the important research fields in education data mining and student modeling.Knowledge tracing is to model the students' knowledge state based on their past exercise records,to track the changes of students' knowledge state as they continue to do the exercises,and then to predict the students' future exercise results.By tracing students' mastery of knowledge skills during learning,more personalized learning guidance can be provided to students.In view of the problems existing in the existing knowledge tracing algorithm,which only updates the students' knowledge status through the results of exercise,considers that there is no difference in learning ability among all students,and that students' learning ability will not change after learning and exercise.This paper proposes two different knowledge tracing algorithms,and designs and implements an online learning prototype system based on these two algorithms.The specific research content and innovations are as follows:(1)A multi-behavior feature knowledge tracing algorithm based on dynamic key-value memory network is proposed.The algorithm first uses the decision tree algorithm to pre-classify the exercise results based on the student's answer behavior data,and then implicitly merges the pre-classification results into a dynamic key-value memory network,fully considering the impact of different learning behaviors on the level of knowledge understanding,This algorithm can more accurately track changes in students' knowledge level and predict future answer results.Experimental results on the ASSISTments 2009 public data set show that the AUC value of the algorithm is 9.2% and 8.1% higher than the current mainstream knowledge tracing algorithms,namely,Deep Knowledge Tracing(DKT)and Dynamic Key-Value Memory Networks(DKVMN).(2)An improved dynamic key-value memory network knowledge tracing algorithm with segmented clustering of learning ability is proposed.The algorithm segmented the students 'long-term exercise records and defined learning ability features,dynamically divided students according to their learning ability,and proposed an exercise record representation method to integrate students' learning ability and exercise-solving behavior.Finally,the reading process of the dynamic key-value memory network is improved,so that it can take into account the students 'current learning ability when predicting the results of students' exercises,and then can predict the results of future exercises more accurately.Experimental results show that the AUC value of the algorithm is 11.4% and 10.3% higher than the DKT and DKVMN algorithms.(3)An online learning prototype system based on a knowledge tracing algorithm is proposed.The core and characteristic functions of the system are intelligent assessment and personalized exercise recommendation for students' knowledge level.After the students complete the exercises,the system can accurately track the students' knowledge level of understanding based on their answer records,and strengthen the weak points of their knowledge points by recommending suitable exercises.To sum up,the algorithm proposed in this paper can more accurately model the students' level of understanding of knowledge skills,and then predict their future exercise-solving results,which is of great significance to help students improve learning efficiency and promote the personalized development of education.