Research On Multi-index Exercise Recommendation Algorithm Based On Students’ Knowledge Tracing

Personalized exercise recommendation is an important topic in the field of educational data mining.At present,many scholars have put forward various exercise recommendation models,but there are still many places that can be improved:(1)The current mainstream models are not thorough in the research of students’ characteristics,and the internal correlation information of students’ overall answer behavior is not fully mined.Students’ learning is a complex behavior,it is necessary to build a high interpretable model.(2)Most traditional models can deal with a limited amount of data,but it’s difficult to deal with the processing of massive online education data.(3)Many models often only consider solving a single problem in the exercise recommendation scenario,but there are more factors need to be considered for a good recommendation behavior.In order to solve the above problems,this paper studies the following three most important indicators in the exercise recommendation system: novelty,difficulty and diversity,and finally forms the main research contents of this paper as follows:(1)In order to realize the novelty index of exercise recommendation,we first try to model students’ learning range of all knowledge concepts and predict the occurrence probability of knowledge points in the next time phase.After that we integrate students’ forgetting regular pattern into the probability prediction model of knowledge points based on LSTM algorithm.Finally we propose the SF-KCAP(Students’ Forgetting Behavior Based Knowledge Concept Appearance Prediction)model according to the above thoughts.(2)In order to meet the condition that the recommended exercises are moderately difficult,we are supposed to predict each student’s mastery of different knowledge concepts,so we study the deep knowledge tracing model DKVMN,which is based on dynamic key value memory networks.After that,we creatively integrate the factors of students’ learning ability and learning attitude into the model to propose DKVMN-CA(Dynamic Key Value Memory Networks with Capacity and Attitude Elements for Knowledge Tracing)model.(3)After modeling the students’ user portrait in the first two points,the diversity index of exercise recommendation is realized by using the collaborative filtering algorithm User CF.Based on the above algorithms,an exercise recommendation system is proposed in this article,which contains two main parts called preliminary screening and secondary filtration,so as to realize the research on multi index exercise recommendation algorithm based on students’ knowledge tracing.(4)The effects of the proposed recommendation model and the DKVMN-CA model involved in the recommendation system are evaluated on four real open-source knowledge tracing datasets,also we compared with other mainstream models to illustrate the advantages of our recommendation system as well as how much it has improved.