Research On Student Learning Behavior Modeling For Online Courses

With the continuous development of science and technology and education,and by the outbreak of the COVID-19 epidemic in early 2020,universities and colleges have adopted online teaching.The scale of online courses is growing.At present,a certain number of students in Colleges and universities have academic failure every year,which seriously affects the development of students,and it is difficult for teachers to track the learning situation of each student.In this context,the research on the modeling method of students’ learning behavior has important practical significance and application value.Students’ learning behavior modeling is to predict students’ future performance according to their historical learning behavior,which plays an auxiliary role in students’ learning.Teachers can also optimize teaching strategies accordingly.There is room for improvement in the traditional modeling methods,which are mainly reflected in: the model structure is simple,and the loss of key behavior characteristics leads to the low accuracy of model prediction;The difference between different students is not taken into account,resulting in the lack of personalized ability;The modeling ability of long series dependence of the model is insufficient.In view of the above problems,this paper has carried out systematic research,and the main contributions are as follows:(1)Most of the existing student learning behavior analysis models are individual prediction models,which are simple in structure and easy to lose key behavior characteristics.Aiming at the problems of low prediction accuracy and poor stability of such models,this paper designs a student learning behavior prediction model based on stacking multi model fusion.First,a two-layer model is constructed.The first layer trains three independent individual prediction models;The second layer uses a linear classifier,introduces the prediction data of the first layer,and obtains the final prediction result through the classification selection of the meta model.Considering the imbalance of student data samples,this paper introduces K-means Smote sample equalization technology to optimize the learning behavior prediction model.Finally,the model is tested on the open data set of ASSISTments online course platform.The experimental results show that the prediction model based on stacking multi model fusion has better performance than the individual model in the evaluation indexes such as accuracy,precision and root mean square error.It is verified that the introduction of K-means Smote sample equalization technology is helpful for the prediction model to effectively screen students with different knowledge levels and teach students according to their aptitude.(2)A depth knowledge tracking method combining the comprehensive difficulty of the problem is designed.In view of the problem that the traditional model does not take into account the differences between different students,resulting in the model can not meet the personalized needs,this paper first preprocesses the topic information.Secondly,based on the grid search method,this paper optimizes the calculation method of the comprehensive difficulty coefficient of the topic,introduces the long-term and short-term memory network model,and introduces the difficulty coefficient of the topic into the knowledge tracking model.Finally,the experiment is carried out on the open data set of the assistments online course platform.The evaluation results show that the model designed in this paper has better prediction effect than the traditional model.The model fully considers the differences between different students,solves the problem of insufficient personalized ability,alleviates the instability of prediction results,and can effectively help students complete their studies and help teachers improve teaching strategies.