| Intelligent tutoring systems based on artificial intelligence technology have a profound impact on the online education industry.The development of online education and the progress of society have also put forward higher requirements for the intelligence and personalization level of intelligent tutoring systems.Knowledge tracing is the task of modeling and predicting students’ knowledge states based on their historical answering interactions on intelligent tutoring systems.Knowledge tracing can effectively predict students’ knowledge mastery and students’ answering correct rate and can personalize students’ answering paths to improve students’ learning efficiency.Knowledge tracing plays an important role in enhancing the intelligence and personalization of intelligent tutoring systems and is a key technology for intelligent tutoring systems to help achieve the goal of smart education.Deep learning has been a great success in artificial intelligence.Various deep learning-based knowledge tracing models have also been greatly developed,excelling in tasks such as knowledge modeling and prediction.However,existing deep knowledge tracing models still generally suffer from the following two problems:(1)One-sidedly focusing on modeling students’ knowledge states based on skill sequences,it is difficult to effectively perceive the association between skills belonging to the same multi-skill composition question.(2)Simply treating students’ historical answering information as serial data and ignoring the graph structure characteristics of the knowledge concepts and skills themselves,and makes it difficult to effectively perceive the association between skills and their neighbors as well as the association between skills in the graph structure.Therefore,this paper focuses on exploring and researching the association between skills and proposes Research on Associated Skills Knowledge Tracing Model Based on Deep Learning.To address problem(1),Associated Skills Knowledge Tracing model is proposed.For multi-skill questions,this paper has systematically analyzed the existing classical datasets of intelligent tutoring systems and discovered new characteristics among the skills of multi-skill questions.This paper defines the conjunctive skills of a question as the current skill and the associated skills.By exploring the potential association between the current skill and the associated skills,the model’s ability to perceive the association between the skills of the same question and the prediction accuracy of the model are both improved,and the depth and breadth of the model’s modeling and prediction of students’ knowledge states are expanded.To address problem(2),this paper proposes Graph-based Associated Skills Knowledge Tracing model.This paper deeply investigates the graph structure properties of skills and knowledge itself,constructs the graph structure of skills based on students’ historical answering sequences,and defines the associated skills in the graph,i.e.,the1-hop neighbors of the skill.The novel combination of the graph structure of skills with graph neural network and graph-based knowledge tracing model improves the model’s ability to perceive the association between skills and their neighbors and to explore the association between skills in the graph structure,which improves the prediction accuracy of the model.In this paper,extensive and systematic experiments are conducted to test the performance of the models proposed in this paper.Neural Turing Machine-Based Associated Skills Knowledge Tracing model achieves the best test AUC results of 85.38%,82.35%,and 80.81% on three real-world datasets(ASSIST09,ASSIST17,EdNet)with question id,respectively.The Graph-based Associated Skills Knowledge Tracing model achieves1.36% average test AUC improvement over the best-performing benchmark models on six chosen public datasets(Algebra05,ASSIST09,ASSIST12,ASSIST15,ASSIST17,EdNet).This paper also explores in depth the detailed process of modeling students’ knowledge states and discusses the tasks such as the association between knowledge concepts,conditional influences between exercises,and knowledge discovery. |
