Research On Student Modeling For Intelligent Tutoring System

With the transition from the Internet era to the age of artificial intelligence,modern education is becoming more intelligent.The way of artificial intelligence + education becomes the development trend of future education.The artificial intelligence education in the big data environment has the characteristics of individuation,precision and comprehensiveness,while reducing the teaching burden of teachers,it has greatly improved the learning benefits of students and promoted the transformation of educational technology.Knowledge tracing,as a method for modeling learner’s knowledge level,plays a key role in intelligent tutoring systems.The research on it is one of the important means to promote the development of education,and has great significance and social value.There are still many shortcomings in the existing student modeling based on knowledge tracing,which are mainly reflected in problems such as insufficient individuation ability,simple model structure,without considering the differences of learners and the relationship between the structure of skills,which ultimately leads to limited prediction of the model and poor interpretability.This paper aims to improve the performance of learner models,from the traditional probability graph models to the latest deep learning-based models,we achieved great improvements in many aspects.There are three main areas of work:(1)Aiming at the problems of insufficient personalization ability and limited prediction accuracy of the traditional Bayesian knowledge tracing model,this paper innovatively introduces evolutionary clustering into the knowledge tracking model.The model clusters the learner’s interaction data in the intelligent tutoring system dynamically over time to obtain student groups of different knowledge levels,then construct an ensemble learning that includes multiple knowledge tracing models.And use an improved algorithm for prediction.At last,this paper proposed an improved strategy for the problem that the EM algorithm is easy to fall into local optimum.The model fully considers the individual differences of the students and the time-smoothing characteristics of the changes in the knowledge level of the learners,which can effectively alleviate the interference of abnormal data and better serve the learners and the teaching system.Moreover,it has better interpretability.(2)Aiming at the problem that the traditional Bayesian knowledge tracing model has a simple structure and it is difficult to use more features for modeling,a knowledge tracing model based on comprehensive difficulty of the problem and input-output hidden Markov model is proposed,which improved the model from the structure aspect.Therefore,when modeling,it is no longer just to use the correctness of the learns’ answers,but to construct different model parameters according to the actual difficulty of the question.Moreover a method for calculating the comprehensive difficulty of the problem combined with response time is proposed.Experimental results show that the model is superior to the traditional model in predicting performance,and superior to other models in scalability.(3)Aiming at the problems of limited prediction accuracy,large volatility and data sparsity of deep knowledge tracing model,An improved deep knowledge tracking model is proposed which consider the structure of the different skills.First,the hierarchical structure relationship and peer relationship between knowledge points are constructed.There must be a similar degree of mastery among the connected knowledge points,and then it is added as a constraint to the loss function of the model.Experiments show that the improved method can further improve the prediction accuracy of the knowledge tracing model and build a better knowledge structure relationship.