Research On Method Of Deep Knowledge Tracing Based On Bayesian Neural Network

In recent years,the online education has been rapidly developed by applications such as education informatization,distance education and web2.0.The current online education is not completely user-centric,and can not provide the most suitable quality education resources for each user according to the user's cognitive level and cognitive style.Guide,so as to teach students in accordance with their aptitude.The prerequisite for teaching students in accordance with their aptitude is to understand the user's cognitive level.Knowledge Tracing(KT)is the de-facto standard for inferring student knowledge from performance data,model students' changing knowledge state during skill acquisition.The existing knowledge tracing models have problems such as incomplete features consideration,poor effect and easy over-fitting in the process of simulating students' knowledge acquisition.In order to better model students' learning and track the process of students' knowledge acquisition,a Bayesian neural network Deep Knowledge Tracing(BDKT)is proposed.This thesis mainly discusses the application of different Bayesian neural networks in deep knowledge tracing.The research work is as follows:(1)In order to avoid the model over-fitting phenomenon and the influence of the few noise data on the model,the Bayesian method and LSTM are combined to make the calculation of the confidence interval of the model possible,and to enhance the generalization ability and anti-noise ability of the model.A deep knowledge tracing method based on Bayes-LSTM is proposed.Experimentally verify the effectiveness of Bayesian neural network in knowledge tracing Introduce student behavior vector learning in the model,and explore the relationship between students' answering behaviors and knowledge points.(2)Aiming at the problems that LSTM structure can not be calculated in parallel and the defects of dealing with long dependence and local feature relationship,it is proposed to replace LSTM with CNN-Attention structure.Combining the Bayesian method with CNN-Attention,a deep knowledge tracing method based on BayesCNN-Attention is proposed to improve the model from the prediction effect and calculation speed.(3)In view of the loss of the relative positional relationship of sequence data by CNN-Attention structure and the lack of long-term dependence of LSTM structure in processing data,it is proposed to replace CNN-Attention with LSTM-Attention structure.Combining the Bayesian method with LSTM-Attention,a deep knowledge tracing method based on Bayes-LSTM-Attention is proposed to further improve the model's ability to fit and long-dependent feature learning.