Research On The Spatio-temporal Distribution And Pattern Evolution Of Learning Behavior In Online Learning

With the widely application of mobile networks and smart terminals,online learning gradually got rid of the restriction of wired networks and fixed terminals,and truly realized the state of learning anytime,anywhere.Students' learning activities also had new characteristics as the learning environment and method changed.In online learning,time and location of learning activities are random and uncertain,and learning strategies will also change with the course progress.In this context,how to identify learning patterns,predict learning performance,and provide instructional assistance required new ideas.Based on student behavior data in online learning,this paper studied the spatio-temporal distribution and evolution characteristics of learning behavior,and then proposed learning performance prediction method based on behavior analysis,and finally provided intervention measures based on the analysis of at-risk students' learning strategies.The research and contributions of this article are summarized as follows:Firstly,this article analyzed the distribution characteristics of students' learning time and location,explored how students learn "anytime,anywhere",and revealed the characteristics of learning performance and group differences.For the dynamic characteristics of students' learning time and location distribution in online learning,this paper proposed the concept of time entropy and location entropy to quantitatively analyze the characteristics of student time and location variation.Research results based on data from a domestic university showed that,there were significant differences in the learning performance among groups with different learning time and location distribution.Students with less learning location types and fixed location had higher course passing rates,while students with more learning time segments and location types had lower course passing rates.Secondly,this article analyzed the students' behavior evolution characteristics and explored the pattern changing laws as the course progresses.In the semester,changes in course requirements caused students to adjust their learning activities and showed evolution characteristics.To analyze the dynamic process of learning patterns,this paper proposed a learning pattern analysis method based on time series feature extraction and segmentation.In the research on K12 data,this article focused on analyzing 11 types of behavioral evolution.The results show that students had various learning patterns in each course stage,and different patterns have positive or negative effects on learning performance."Balanced","Early Bird" and"Interaction" were the patterns that had positive effect on academic performance.;"Oscillative","Low-Engaged" and "Low-Interaction" were patterns that had negative effect on learning performance;there were obvious performance differences among behavioral evolution types,EB-B-B students achieved the highest passing rate,I-O-LE students achieved the lowest passing rate,and most students with high learning performance have a higher learning frequency or consistent learning pattern than students with low learning performance.Finally,this article researched learning performance prediction and instructional intervention based on the analysis of learning behavior,identified at-risk students,and proposed intervention measures.For the problem of identifying as many at-risk students as possible,advancing intervention time and providing individual intervention measures,this paper proposed an early warning model based on LSTM-autoencoder feature extraction and attention weight calculation.Research results based on K12 data showed that,the proposed method improved the recognition rate of at-risk students and advanced the intervention time point.This method also had obvious advantages in identifying low-interaction and non-persistent at-risk students.The analysis of students' patterns and learning strategies showed that the essential difference between the success and at-risk students was that,due to at-risk students' "holiday effect",they cannot form a complete feedback loop of self-regulated learning to adjust their learning activities.This article proposed intervention measures for at-risk students from two perspectives—supplemented the learning activities that they lacked according to the SRL model,or guided the at-risk pattern students to change to a similar success pattern.This article focused on the three core issues of behavior pattern recognition,learning performance prediction and instructional intervention,revealed the law of learning,and had achieved specific results.However,due to the limitations of the data and methods,the work of this article still has its shortcomings,which need to be further improved.In terms of data,it is necessary to integrate multi-source and multi-modal data to describe the students' learning behavior more completely.In terms of models,deep learning model architectures are constantly being introduced.Whether there is a model that can further improve the results of learning data analysis still needs to be explored.The future work will focus on improving the results from the perspective of data and method,and expanding the instructional application of learning pattern analysis.