Research On Deep Knowledge Tracing Model Based On Q Matrix

Purpose —— The online education industry is in full swing,which has also put forward higher requirements for online education.Online education is developing towards the direction of personalization and intelligence.Knowledge tracing is an important method to achieve the personalization of online education by using artificial intelligence.It uses learners' personal learning data to diagnose and predict learners' learning status and provides personal guidance for learners.At present,although the research of knowledge tracing model has made great progress,there are still some problems,such as the accuracy of the prediction results is not enough,the convergence speed is not fast enough and problems about multi knowledge point processing methods.These problems limit the use and promotion of knowledge tracing model to a certain extent.Therefore,it is of great significance to study the knowledge tracing model and improve its performance to promote the development of online education.Method —— This paper mainly studied the deep knowledge tracing model based on Q matrix.Firstly,we have used the knowledge points in the education data which were marked by experts to mark information,that is,several knowledge points contained in each exercise,to establish the corresponding relationship between exercise and knowledge points and construct Q matrix which will be used to calculate correlations between exercises.Correlation degree can be used to calculate the probability of other related exercises answering right or wrong when a student has answered a certain exercise right or wrong,and can be used to process the student's study data.After the processed data was imputed into the deep knowledge tracing model to train it.At present,the common practice of deep knowledge tracing model is to input the right and wrong information of a student answering a certain exercise at a certain time in a single training.In this paper,the input of deep knowledge tracing model based on Q matrix includes not only the right and wrong information of students answering a certain exercise at a certain time,but also the predicted probabilities of other exercises that have certain correlations with this exercise will be answered right or wrong.Finally,the model was used to predict the students' right or wrong situation of the corresponding exercises and judge the students' mastery of the exercises.Research Results —— This paper designed the experiment plan and carries out the experiment on the public data set,according to the research objectives and methods.Experiments have showed that the deep knowledge tracing model based on Q matrix combines the characteristics and advantages of traditional cognitive diagnosis model and knowledge tracing model.Compared with a single cognitive diagnosis model,the model avoids the interference of human factors to a certain extent through deep learning,and improves the accuracy of experimental prediction.In addition,compared with the knowledge tracing model without Q-matrix,this paper introduced the corresponding relationship data between exercises and knowledge points which added the correlation information between exercises into the experiments and solved the problem of multi knowledge point processing to a certain extent.These methods effectively speeded up the convergence speed of the model and improved the performance of the deep knowledge tracing model.Limitations of research —— The Q matrix constructed in this paper needs to use the mapping between exercise and knowledge point,that is,an exercise contains one or several knowledge points.Therefore,this kind of information must be included in the experimental data set and this method of this paper cannot be used for the exercises data without such annotation.In addition,the relationships mapping information is usually manually annotated by domain experts which may has strong subjective dependence.The accuracy and quality of the annotation information will also affect the final prediction results.Actual impact —— This paper studied the deep knowledge tracing model based on Qmatrix.The mapping relationships between exercises and knowledge points were processed by using Q-matrix,and they were used as the input of deep knowledge tracking model.It combined the advantages of traditional cognitive diagnosis model and deep knowledge tracing,and improved the performance of the model.Originality —— The Q-matrix was constructed by using the mapping between exercises and knowledge points marked by experts,and the correlations between exercises were calculated by using the Q-matrix.The input of the deep knowledge tracking model was not only the result of a single exercise at a certain time,but also the prediction probability of all other exercises that have reached a certain degree of correlation with the exercise.This method improved the accuracy of the prediction results of the knowledge tracking model,speeded up the convergence speed of the model,and solved the problem of multi knowledge point processing to a certain extent.