Graph Embedding And Double-ended Attention Based Knowledge Tracing And Learning Path Recommendation

Knowledge Tracing is a technology that used to evaluate students' mastery of knowl-edge skills,and to predict whether students can answer specific exercises correctly.In recent years,with the continuous increase of online education platforms,the demand for personalized learning has become more and more urgent.Knowledge tracing can trace the knowledge states of learners,and learning path recommendation can carry out person-alized planning and recommendation for the learning process of learners according to the prediction results of knowledge tracing to meet the learning needs of each learner.Although the research on knowledge tracing and learning path recommendation keep advancing in recent years,there are still following problems and challenges need to be solved:(1)The problems from the online education platform cannot be simply embedded for its high dimensionality and sparsity,and using knowledge skills instead of questions will cause information loss.(2)The existing knowledge tracing methods lack the ability to analyze and represent learner's knowledge state,as well as the participation of attention mechanism.(3)The traditional recommendation method lacks the consideration of effec-tiveness of improving the knowledge mastery and ability of the learners.Based on the above challenges,this paper conducts in-depth research on graph embedding,knowledge tracing,and contributes the following works:First,for the representation of high-dimensional sparse question features,we pro-posed a novel graph embedding method to represent the latent correlation between ques-tions based on a weighted graph,which solves the problem of high-dimensional sparse fea-ture representation.Second,for the application of the attention mechanism in the knowl-edge tracing model,we proposed a Student attention-based and Question-aware model for Knowledge Tracing which add attention mechanism to both input end and prediction end.The model uses a memory matrix to represent the learning state of students that no longer limited by the number of neurons and parameters in the neural network,which im-proves the prediction accuracy of the knowledge tracing model.Last but not least,for the knowledge level improvement of learners in learning path recommendation,we proposed a learning path recommendation algorithm based on knowledge tracking model.Use the graph embedding method and knowledge tracing model to represent the question features and recommend based on Item-CF,to improve the learner's knowledge level.Sufficient experiments are conducted on four real world data sets.Through control,ablation,and simulation experiments,methods in this paper are proved to be effective and reasonable.