Research On Personalized Recommendation Of Mathematical Knowledge Based On Knowledge Tracin

With the rapid rise of online education,a huge amount of educational data have been generated,and it is particularly important to mine and analyze the personalized information implied in the data to optimize the quality of teaching.In the face of massive learning resources,it is difficult for students to accurately learn the required knowledge in the limited period of time.They may repeat learning what they have already mastered or spend inordinate amounts of time learning knowledge that is too difficult for them.Especially for elementary and secondary school students with significant difference in cognitive structure and level,it is more necessary to gradually learn knowledge and skills that consistent with their cognitive level according to their own differences.As an important research direction in the field of educational data mining,Knowledge Tracing(KT)task can accurately model knowledge state of learners based on historical learning data and predict their future performance in answering questions.However,the existing KT methods with better performance like dynamic key-value memory network(DKVMN)use the correctness of the exercise as the only basis for modeling knowledge mastery degree,failing to consider the influence of forgetting and behavior features of learners and features of exercises on modeling results.At the same time,in the previous mathematics learning resource recommendation research,researchers mostly focused on single resource recommendation for exercise or knowledge points,and rarely considered joint recommendation from the two levels.To solve the above problems,this thesis proposes a personalized recommendation method for mathematical knowledge based on knowledge tracing.The specific research contents mainly include the following two aspects:(1)Aiming at the problem that DKVMN model fails to consider the influence of the knowledge forgetting phenomenon of learners,learning behavior and the features of exercises on knowledge tracing results,the model called dynamic key-value memory network by considering forgetting mechanism and multiple features(DKVMN-FMF)is proposed.In order to fully take the influence of multiple features on knowledge mastery into consideration,DKVMN model is firstly integrated with different features including the number of attempts,hint request,duration time,the type and difficulty of exercise.At the same time,we further measure the different importance of the features for improving the performance of the model by using attention mechanism to assign different weights to them.This model adopts power function that decays over time of fitted forgetting curve to model the forgetting behavior of students and gain a more comprehensive and accurate knowledge level of students.The experimental results on open datasets for knowledge tracing called ASSISTments2009 and ASSISTments2012 show that the prediction performance of the extended model is superior to the original model and two existing advanced methods in terms of AUC and accuracy.(2)Considering that the existing researches on the recommendation of mathematical knowledge less take both knowledge points and exercise recommendation into account,a personalized recommendation method for mathematical knowledge based on knowledge tracing(KT-PRMK)is proposed.Firstly,we obtain the knowledge mastery degree of students according to the proposed knowledge tracing method and divide the mastery stages into three parts.This method recommends knowledge points or exercises that meet students’ cognitive level for different stages.For knowledge points that have not been mastered or have been mastered,recommend the predecessor or successor knowledge points to the student based on the semantic relationships between knowledge points in the built mathematical knowledge graph.For knowledge points that are weakly mastered,recommend the similar exercises to the student according to the recommendation algorithm.At the same time,the exercises that may be of interest to students are also recommended to them from the preference perspective.The experimental results show the effectiveness of this recommendation method in personalized recommendation of mathematical knowledge.