Research And Application On Knowledge Tracing Model Based On Graph Embedding And Attention Mechanism

With the rapid progress of technology and the promotion of education informatization,online learning has become one of the mainstream learning methods nowadays.However,students’ online learning effect is much lower than offline.To improve students’ online learning effectiveness,it is necessary to realize personalized teaching for students,targeting each student’s own learning characteristics and formulating a learning plan that suits them.Therefore,how to scientifically analyze and assess students’ mastery of knowledge concepts based on their historical learning records,to obtain detailed knowledge levels of students has become a hot research topic in the field of education technology nowadays.Knowledge Tracing(KT),which can predict the future learning performance of students based on their current learning performance and thus obtain their mastery of knowledge concepts,is one of the core issues in achieving personalized education.Researchers have extensively combined KT with machine learning algorithms and achieved some results.However,KT still faces many problems because of the complexity of students’ online learning behavior.Students’ online learning data is very rich and multidimensional,and existing KT methods are too one-sided in collecting data,leading to sparsity of features.Moreover,students’ knowledge level is a dynamic process of change,and many existing KT models do not make good use of the time factor and do not better reflect the impact of forgetting behavior on students’ knowledge level.To address the above problems,this thesis proposes two KT models based on neural network algorithms such as Graph Embedding and Attention Mechanism,consider the potential correlation between students’ multidimensional learning features and introducing the time interval factor.We also design and implement a learning analysis system based on KT.The main work and innovations of this thesis are as follows:(1)To address the feature sparsity problem of existing KT models,this thesis proposes a KT model based on Feature Potential Correlation Graph Embedding(FPCGE).First,the embedding is obtained based on students’ learning behavior feature data using Node2 vec algorithm,and then,the importance of different features on students’ knowledge level is obtained based on Graph Convolutional Networks(GCN).Finally,student performance is predicted based on Long-Short Term Memory(LSTM)network.Numerous experimental results show that the FPCGE-based KT model can effectively alleviate the feature sparsity problem and improve the accuracy of KT prediction results.(2)To address the problem that the existing KT models do not trace students’ mastery of knowledge concepts with sufficient accuracy,and cannot model the dynamic change process of students’ knowledge level well.In this thesis,based on FPCGE,we propose a Graph-Attention Knowledge Tracing(GAKT)model based on graph embedding and multilayer attention network by introducing time interval factor and attention mechanism.Firstly,the question aggregation embedding is obtained based on FPCGE,and the time embedding is obtained based on One-hot Encoding according to the value of the time interval of students’ practice questions,and the two embeddings are spliced and input to the multilayer attention network.Next,a multilayer attention network consisting of a question attention network focusing on question features and a time attention network focusing on time features is constructed.Finally,the output between each attention layer is adaptively fused by a gated fusion unit to predict student performance.Experimental results on multiple datasets show that the model outperforms multiple comparison methods.(3)A learning analysis system based on KT is designed and implemented.Based on the results of KT,students’ mastery of knowledge concepts is analyzed,and students’ mastery of knowledge concepts is presented in a more visual and detailed manner,which is a guide for teachers to provide learning suggestions to students and develop personalized learning programs.