Analysis Of Students’ Learning Effect Based On Online Behavior

With the rapid development of online education,educational data mining(EDM)has become a research hotspot.With these research results,online education platforms can provide help for improving online education,optimizing classrooms,and providing personalized education.In order to study online education and the student’s learning effect(that is academic performance),we conduct research from two aspects.(1)Knowledge tracing is a common method to evaluate the student’s learning effect.It also becomes more popular in the research of personalized education.It evaluates students’ learning effects by using the students’ answers to the exercises.However,forgetting is unavoidably in the learning process,so we propose a dynamic key-value memory network combined with forgetting-curves(DKVMN-FGC)to analyze and compare students’ knowledge proficiency.In this model,we simulate forgetting during learning by adding a forgetting curve to the “reading process” of the original model.Our experiments were conducted on two datasets of different learning domains.Furthermore,we compared our model with the other three existing models.The results prove that our improved model can effectively evaluate students’ knowledge mastery,and the area under the receiver operating characteristic curve(AUC value)and the accuracy rate are significantly higher than the other three models.Then,according to the experiment,we obtained the potential knowledge components of the exercises,furthermore,we obtained the students’ mastery of knowledge points over time.(2)In addition to analyzing the students’ exercises,we can also start with factors that affect students’ performance.And we can propose learning rules for these key factors to help both teachers and students react in advance,thereby helping to improve student performance and achieve personalized education.We use the distributed CART decision tree algorithm in spark to establish rules for real data sets and find out relevant factors.The experimental results show that when the score is only divided into pass or fail,the model obtains 20 learning rules with an accuracy rate of 78.34%;when the score is divided into three levels of low,medium,and high,the model obtains 12 learning rules with an accuracy rate 76.38%.In both cases,“age” is the most important factor,and other relevant factors include “the entry time of students”,“age”,“total number of video clicks per course”,“gender”,“course id(different courses)” and so on.At the same time,we propose some suggestions according to the experimental results.