Research On The Model For Predicting At-risk Students In Online Course With Fusion Of Advanced Behavior Characteristics

Online education is booming,and online learning has become the educational norm alongside offline teaching.The number of online learners is increasing,but the passing rate of online learning courses and the number of learners who have obtained certificates have decreased.High registration rate,low participation rate and high dropout rate have become one of the problems that need to be solved urgently in online learning.At present,relying on big data technology,machine learning methods and so on,if we can predict learners' learning choices in the next step in real time according to a large number of learners' existing online learning behaviors,and give timely warning to learners facing learning difficulties or dropping out of classes,teachers can make timely intervention measures to help them complete the course smoothly.At present,there are researches on learning early warning and learning outcome prediction at home and abroad.Most of them choose the underlying behavior characteristics which are closely related to learning outcomes for learning prediction,and give early warning to learners who may be in learning crisis.The experimental data used in the research come from different learning platforms.Because there is no unified learning behavior standard,the behavior data collected and recorded by each learning platform are different.At the same time,most of the research uses the basic behavior characteristics collected by the platform,and does not mine the advanced behavior characteristics,so the effect of learning early warning is not optimal.Because learning is a dynamic process which is constantly changing,and it is also a complex process which is influenced by many factors and works together,it is expected to mine and extract learning characteristics from many aspects and in-depth,so as to comprehensively and deeply depict the whole learning process and provide real-time and accurate learning warning.In this paper,based on the basic behavior characteristics of human-computer interaction,mining and extracting the time-sequence characteristics and time-varying characteristics of human-computer interaction,we construct the learning early warning model integrating the time-sequence characteristics of human-computer interaction and the learning early warning model integrating the time-varying characteristics of human-computer interaction respectively.Based on the basic behavior characteristics of human-human interaction,we mine and extract the semantic characteristics and network characteristics of human-human interaction.From the perspective of teaching and learning,this paper constructs an integrated learning early warning framework with advanced behavior characteristics from multiple aspects.The main innovations of this paper mainly include the following four aspects:(1)We construct a learning early warning model which integrates the time-sequence characteristics of human-computer interaction behavior.This paper focuses on the influence of the sequence of learning behavior on the learning results.Based on the self-regulated learning theory and the sequential pattern mining,we propose an online learning early warning model integrating the temporal characteristics of human-computer interaction.The experiments show the overall accuracy of the four online learning prediction models and the accuracy of identifying dropouts,recall rate and F1 value are improved to a certain extent after integrating the time-sequence characteristics of human-computer interaction behavior into the basic characteristics of human-computer interaction behavior.This shows that the time-sequence characteristics of human-computer interaction can not only improve the overall accuracy of online learning early warning model,but also improve its performance of identifying dropouts.(2)We construct a learning early warning model which integrates the time-varying characteristics of human-computer interaction behavior.The existing researches on learning early warning mainly focus on the cumulative number and frequency of human-computer interaction behaviors,ignoring the impact of the dynamic change of the number of interaction behaviors over time on learning outcomes.This paper focuses on the time-varying characteristics of human-computer interaction,combined with the memory function of recurrent neural network,on the basis of the basic characteristics of human-computer interaction collected by the existing common learning platform,integrates the time-varying characteristics of human-computer interaction,and constructs an online learning early warning model based on LSTM network and integrating the time-varying characteristics of human-computer interaction.Comparing the learning early warning model proposed in this paper with the traditional early warning model based on machine learning,the experimental results show that the time-varying characteristics of human-computer interaction behavior can better predict the learning results,and the experimental results show that the prediction effect of LSTM learning early warning model with time-varying characteristics is significantly better than the traditional early warning model based on machine learning.(3)We construct a learning early warning model integrating the semantic characteristics and network characteristics of human-human interaction behavior.Using natural language processing technology and semantic analysis technology,according to the forum posts and knowledge content,interaction depth,extract the semantic characteristics of human-human interaction;using social network analysis technology,combined with the individual importance of learners in the forum,extract the network characteristics of human-human interaction.Based on the basic characteristics of human-human interaction behavior,the semantic characteristics and network characteristics of uman-human interaction are integrated in turn respectively.The experiments show that based on the basic characteristics of human-human interaction,with the integration of semantic characteristics and network characteristics of human-human interaction,the overall AUC of multiple learning early warning models is improving,and the performance of identifying dropouts is improving.The AUC of learning early warning model integrated with semantic characteristics and network characteristics is the highest,and the performance of identifying dropouts is the best.(4)We propose an integrated learning early warning framework integrating advanced behavior characteristics.The integrated learning early warning framework proposed in this paper has two advantages:first,it makes use of many advanced r behavior characteristics,such as the time-sequence characteristics and time-varying characteristics of human-computer interaction behavior,the semantic characteristics and network characteristics of human-human interaction behavior,which helps to improve the accuracy of each basic prediction model;second,it adopts a dynamic selective integrated learning mechanism to ensure the accuracy of each selected basic prediction model,the output of the basic prediction model is effective,which helps to improve the accuracy of the final prediction results.