Design Of Intelligent Tutoring Algorithm Based On Knowledge Tracing

The intelligent education system can establish a learner-centered education environment and improve students'learning efficiency.It has been included in the national new generation artificial intelligence development plan.The intelligent guidance system is able to track the mastery of each knowledge concept of students through student exercises,and adaptively recommend relevant exercises to students to enhance students'mastery of knowledge and improve students'learning efficiency.The research on it has important theoretical and practical significance.At present,there are two bottlenecks that restrict the performance of the intelligent guidance system:1)The knowledge tracing model cannot accurately track the knowledge status of students and predict the results of student exercises.The reason is that the current knowledge tracing model does not have a reasonable model structure to effectively use the exercise's concept features(knowledge point labels),resulting in poor model performance.2)The design of the exercise recommendation algorithm relies on manual rules is not efficient.The heuristic-based exercise recommendation algorithm only focuses on the students' short-term performance improvement,and it is difficult to find exercises that make students' ability to improve steadily.In view of the above problems,this paper is based on a large-scale,real online education system's exercise database and students practice data,measures and observes the effective features related to the students'answering results,designs the effective representation of features and the knowledge tracing model which uses the new knowledge concept structure.This paper creatively applies deep reinforcement learning to exercise recommendation,designes exercise recommendation policy algorithm,and evaluated the policy based on real data.The specific contributions are as follows:(1)In order to improve the performance of the knowledge tracing model,the hierarchical structure of the exercise concept is innovatively introduced into the design of the neural network,and a new deep learning knowledge tracing model is designed to improve the model tracing performance.Firstly,based on the multi-level knowledge concept feature of the exercise and the dynamic key-value memory network DKVMN,a concept-aware knowledge tracing model DKVMN-CA is proposed.The experiment proves that the neural network structure of the model can effectively utilize the multi-level knowledge concept features of the exercise for knowledge tracing,which significantly improves the performance of the knowledge tracing model and has 1.2%of AUC higher than DKVMN.Secondly,the model is modified by adding features such as difficulty,stage features,and the duration of finishing exercise,which further improves the performance of the knowledge tracing model and has 1.9%of AUC higher than DKVMN.(2)With the improved knowledge tracing model DKVMN-CA as the student simulator,innovatively introduce deep reinforcement learning into the exercise recommendation algorithm to optimize the exercise recommendation policy.The exercise recommendation policy can be based on the student's history of doing exercises,considering the long-term performance improvement of the students to recommend exercises,in order to maximize the degree of knowledge of the students after completing the exercise recommendation sequence.The experiment proves that the algorithm can find the exercises that improve the students'grades in order to continuously improve the students'knowledge level,and solve the traditional heuristic recommendation algorithm problem that after a certain number of exercises,the exercise can not be found to raise the student's grades.To the best of our knowledge,this is the first time that deep reinforcement learning has been applied to the math exercise recommendation,providing a new reference for future exercise recommendation methods.Figure 17,table 2,reference 46.