Reaserch On Deep Knowledge Tracking Model Based On Graph Embedding

With the widespread use of the Internet,online education is also booming.Online education has gradually become an indispensable part of the field of modern education system due to its learning resources are rich,high-quality and real-time updating,and the learning methods are flexible,convenient and free of time and space constraints.At present,with the breakthrough of artificial intelligence technology,online education has gradually transformed from traditional platform resource education to intelligent education.Personalization is the core of the intelligent education system,and if you want to achieve personalized education,you must understand the students’ real-time knowledge level.Knowledge tracking is a means to model the entire learning state based on the learner’s historical learning trajectory to predict the learner’s future performance.Through knowledge tracking,you can grasp the student’s learning state in time,so you can recommend relevant exercises to them in a targeted manner,so as to teach students according to their aptitude,and improve learning efficiency.However,for knowledge tracking,researchers mostly focus on the model itself,and ignore the relationship information between topics.Based on this,this paper proposes a deep knowledge tracking model based on graph embedding.By treating the exercises as nodes and the relationships between the exercises as edges,a graph network based on statistical information similarity or semantic similarity is applied to the exercises.After construction,two representative graph embedding algorithms LINE and Node2 vec are used to embed the graph network of the exercises,generate a low-dimensional dense vector containing the relationship information between the exercises and the exercises,and use this as the input of the knowledge tracking model.In the knowledge tracking model,based on the traditional deep knowledge tracking model,we set an embedding layer for the input of the graph embedding vectors,so that the model can use the related information between the exercises during the learning process.In the circular layer of the deep knowledge tracking model,we respectively use the GNN and LSTM which are variants of RNN,and also combining the attention mechanism to control the model’s ability to learn important information.In the output layer of the model,we set up a specific output structure,ahead from the idea of Laplace feature mapping,and adding a graph loss term based on the exercises similarity matrix to the loss function of the model,thereby further improving the model to use relationships of exercises.Through experiments on real data,it proves the rationality and effectiveness of our proposed deep knowledge tracking model based on graph embedding,and through a series of model improvement comparison experiments,we verify the feasibility of our improved method.In the final model on the output,its AUC is stable higher than the traditional deep knowledge tracking model.