Research On Deep Knowledge Tracking Based On Attention Mechanism

With the rapid development of the information age,the education field has gradually become intelligent,and many online education platforms have emerged,such as: ASSISTments,ed X,MOOC and Zhixue.com.Due to the popularity of online education platforms,a large amount of learning data is generated.Researchers hope to track every student’s knowledge state through a large number of data generated by online education platform,and provide personalized learning guidance to students.Knowledge tracking methods can effectively achieve this research goal.Regarding knowledge tracking methods,researchers have made a lot of research.There are currently widely used methods: Bayesian knowledge tracking model,matrix decomposition model,tensor factor decomposition method,Knowledge tracking model based on deep learning,etc.At present,the most effective method is a knowledge tracking model based on deep learning,that is,a deep knowledge tracking model,which uses a recurrent neural network to use long and short-term memory models to track students’ knowledge mastering status over time.The deep knowledge tracking model only uses the student number,question number and correct answer information recorded by the students in the online tutoring system.However,relying on the correct answer information to network modeling is easy to lose important information and there is no practical theoretical interpretability,so in this study,based on the deep knowledge tracking model,the hidden feature information contained in the online learning records is fused to improve the performance and interpretability of the model.The innovative results achieved in this article are as follows:(1)A deep knowledge tracking model based on knowledge point text is proposed.The model uses a small amount of conceptual information in each topic to extract the similarity between each topic,calculating the attention vector of each student’s exercise sequence,combining it with the long and short-term memory model,and improving the deep knowledge tracking model to improve the ability to predict students’ academic performance.Experimental results show that this method has a certain improvement in AUC and ACC.(2)A deep knowledge tracking model based on multi-feature attention mechanism is proposed.The model extracts multiple hidden features contained in the online tutoring system based on student practicerecords,for example: according to the question number and sequence number of each student at a certain time,calculating the number and accuracy of the exercises before this time,as well as the sequence number of the subject in the previous exercise to obtain the sequence interval between the two questions,and the attention vector obtained after fusing these three features is combined with the deep knowledge tracking model.Experimental results show that the attention mechanism based on multi-feature fusion has a further improvement in AUC and ACC compared to a single feature.The study improves the performance of the deep knowledge tracking model to a certain extent,and provides technical support for the online tutoring system to provide personalized tutoring and recommend learning resources to students.