With the boom in education and digital technology,large amounts of student learning data are beginning to be stored in electronic form within universities.Predicting student performance(e.g.,grades)in future studies has become an important topic.In reported research on student performance prediction,most of them overlook the implicit correlation among students'historical exercises.Therefore,we propose a method to predict student performance based on knowledge networks and knowledge tracking.The main idea is to integrate the knowledge tracking and the knowledge network formed by the association between knowledge points of exercise.The pro-posed method takes the knowledge network composed of exercise-specific knowledge points and their precursor knowledge points as one of the inputs and masks the knowledge points with mastery level greater than a given threshold in the knowledge network.Combined with LSTM,students'knowledge level is tracked through the integration of learning records and attention mechanisms,ultimately predicting students'performance in future problem-solving.The experimental results on public datasets show that the model proposed improves by 2.3％,2.2％,3.3％,3.4％,0.41％and 2.0％on the Assistment2009 dataset,and by 9.0％,9.4％,9.7％,9.3％,8.1％and 8.5％on the Assistment2015 dataset,respectively,compared to DKT,DKT+,DKVMN,SAKT,MFKT and MSKT,has better model accuracy.
维普
万方数据