Research On Knowledge Tracking Algorithm Based On Recurrent Neural Network And Factorization Machine

Knowledge tracking refers to tracking students’ learning status according to their historical answers.Its significance lies in discovering students’ mastery of knowledge concepts and predicting students’ future performance.In recent years,with the rise of online education,the original offline teaching mode gradually shifts to online education.This way can break through the time and space constraints of traditional education and improve the learning efficiency.it can also make the average education resources and avoid the problem of uneven distribution of educational resources caused by regional environment.As more and more online education institutions emerge,some problems are highlighted.Many online education institutions simply move materials from books to the Internet,It will make students get half the result with twice the effort and students’mastery of knowledge concepts is often difficult to reflect through online education.Therefore,knowledge tracking has gradually become an important research topic in the field of online education and computer aided education.There are three classic knowledge tracking models:Bayesian knowledge Tracking,Deep knowledge Tracking and Deep Knowledge Tracing with Dynamic Student Classification.All of these three models have the problem of single input vector,that is,only the right or wrong questions and the question number are taken as the input vector.At the same time,these models do not distinguish the input of different moments,thus neglecting the influence of key questions in the answer sequence.Based on the mentioned disadvantages of the above algorithms,this thesis puts forward a new knowledge tracking model.We combine the attention mechanism in the Recurrent Neural Network with the Factorizer Machine in the recommendation algorithm.The attention mechanism is used to mine the key points in the input vector at different times.These input vectors are made up of students answer condition,so the user can grasp the important knowledge points.FM is used to find out the interactions between features,so that the influence of environmental factors which are other than the question number on students’ answers can be taken into account.The RNN-FM algorithm proposed in this thesis effectively avoids the single input problem in the traditional knowledge tracking algorithm,and it can also carry out personalized answer exercises for students according to the key answer moments.The data set used in this paper is the student answer data provided by the Assistments intelligent guidance platform,which is widely used in educational scenarios such as knowledge tracking.The version used in this paper is Assistment-2009.The data set contains 525,534 student answer records,with a total of 4217 students and 123 knowledge points,and each record contains 30 features.The new model is experimented on different training sets and test sets.The accuracy rate of students’ answers on the test is quantitatively analyzed,and compared with the classical knowledge tracking algorithm.The experimental results show that in the data set with a large number of features and long answer sequence,the new model can better simulate students’ mastery of knowledge points,and perform well.