A Deep Learning Knowledge Tracing Model Based On Attention Mechanism

E-learning is the product of the combination of Internet technology and computer technology with educational needs.In recent years,it has received more and more attention and vigorous promotion from education circles.On the premise of good infrastructure conditions,compared with traditional education forms,online education platforms are user-friendly learning time and learning locations,large-scale teaching costs are lower,and various electronic or multimedia teaching can be connected more smoothly Resources and in some aspects can more conveniently record learner’s behavior and feedback and other advantages.However,due to its large-scale and online nature,online education platforms are lacking in personalized guidance to students.Therefore,the establishment of an intelligent learning guidance system has become an important research topic in online education.Knowledge tracing is a task of modeling the degree of mastery of students’ knowledge components based on historical practice records,and is an important research on the learner model of the intelligent guidance system.The basis of the research is the machine learning method and the large number of student course learning history records provided by the online course platform,which belong to the data mining task in the education field.Early knowledge tracing methods are often based on hidden Markov models;after the rise of deep learning,the structure based on recurrent neural networks has attracted the attention of knowledge tracing researchers;but the more novel methods are based on attention mechanism models.In the field of sequence modeling in recent years,a variety of methods based on attention mechanisms have emerged,and one of the important deep attention neural networks is the Transformer structure.This paper proposes a method for knowledge tracing based on Transformer,and carries out two researches on the realization of model and transfer learning for knowledge tracing task:(1)A deep knowledge tracing model based entirely on the attention mechanism is designed.Compared with some previous models based on the attention mechanism,the method of this topic improves the embedded representation of interactive records,designs a gate structure suitable for this method,and optimizes the input processing of the self-attention sub-layer to improve in-depth knowledge Track the predictive performance of the model.The experimental results on four commonly used public data sets for knowledge tracing show that compared with previous methods,the model proposed in this paper can better reflect learners’ mastery of knowledge points,and it is better on data sets with large sample sizes.good performance.(2)The knowledge component elements of the knowledge tracing data set are analyzed through co-occurrence associations and manual annotation methods are used to construct a knowledge graph,and a vector is generated using the method of graph embedding representation.This vector can be used in a variety of knowledge tracing models that use embedded representations as input,and participate in model training in the form of solidified or non-solidified coding layer parameters.According to the experimental results,using this method compared with the method of randomly initializing the coding layer parameters,it has obtained a 2% improvement on two different models,which shows the effectiveness of the pretraining method proposed in this paper.