Student Model Of Adaptive Learning Based On Deep Learning

With the development of economy,people pay more and more attention to the education.In the field of education,it is considered that "one to one" teaching pattern is most efficient,but there are not enough teachers to teach the lager student.How to solve this problem through the Internet and AI is a very hot research fields.Traditional off-line teaching organization such as New Oriental?Tomorrow Advancing Life?HuJiang and so on apply the "one to one" teaching pattern.This pattern can get a good efficient,but it is heavily dependent on the level of the teacher's teaching.Teachers are scarce resources,so it could not maximize the satisfaction of all students.Another teaching pattern which called MOOK,put the teacher's class video recorded on the Internet,students can study at any time,and can learn from different schools and teachers in the world.But MOOK pattern requires students to have a very strong self-control ability.And when students encounter problems without corresponding teachers for eliminating confusion will result in the loss of the serious problem of online students,research shows that only 20%of students will complete a course of study.So if we can put a line under the advantages of one to one teaching mode and advantages of Internet MOOK form together,it can greatly improve the quality of teaching.Online teaching platform integrate teaching resources through the Internet.It can provide a large number of high-quality teaching resources,so anyone can study through the Internet without obstacle,which is convenient for the majority of peoples.However,the traditional Online teaching platform does not know the situation of students' mastery of the knowledge,so it could not achieve personalized adaptive teaching.At the same time,due to the poor self-control of students themselves and low learning efficiency,resulting in the loss of online personnel,only a small number of students can learn a course completely.Adaptive learning evaluating the situation of students' knowledge in real time,so as to realize the adaptive personalized teaching.It can improve the students' learning efficiency and reduce the student turnover rate.Therefore,how to estimate a student's proficiency based on their previous inter-action is a key challenge for the adaptive learning.Do exercise is very important for students after they leaning some courses.It is also a very necessary feedback form for teachers which can improve their teaching methods and content.In this paper,a brief description of the exists problems about the Online teaching platform currently.Introduce the research status of the adaptive learning at home and abroad,especially the student model.Now,the commonly student model includes Item Response Theory(IRT)?Bayesian Knowledge Tracing(BKT)and Deep Knowledge Tracing(DKT).Then a series of general description of the current development of the deep learning technology and some popular models are given.The model includes Convolutional Neural Network(CNN)?Deep Belief Networks(DBNs)and Recurrent Neural Network(RNN).Last we apply DKT to estimate student proficiency base on the RNN and tracing the knowledge based on their previous inter-actions with the system.Then we compare those student model.For the model of DKT,we apply Long Short Term Memory NetWork(LSTM)which is the improved model of RNN.Also we apply some mechanism to improve the performance of DKT model such as Attention?RNN Bucketing and multilayer RNN model.The experiments indicate that DKT can work effectively comparing with BKT and IRT,and those mechanism can improve the DKT's performance in accuracy.auc and so on.