Research On Data Analysis And Mining About MOOC Learners

In recent years,Massive Open Online Course(MOOC)has been developing rapidly,causing a storm of educational technology revolution.The number of learners in MOOC is huge,and they have produced a lot of data during the process of MOOC studying.Through the analysis and excavation of these data,it can bring practical significance to the MOOC learners,teachers,and the development of MOOC.Therefore,relevant researches are gradually becoming the topic that researchers pay attention to.This thesis takes MOOC learners as the entry point of research,and mainly carries on the following aspects:(1)This thesis explores the factors that influence the learning results of MOOC,from qualitative(visualization)aspect and quantitative aspect(statistical analysis,correlation test),and provides important reasons for feature selection.In the process of analysis,the "outlier" learners are found.Their learning behaviors are different from general learners.Finally,from the macroscopic and microcosmic angles,the law of analysis is summarized.(2)A large number of studies have simply analyzed the phenomenon of outlier learners and have not been studied in depth.Thus,this thesis presents an outlier learner detection model based on density clustering to comprehensively recognize them.The learner's learning behavior and learning result are regarded as cluster variables,and the feature-weighted algorithm is used to relate the relationship between variables and learning results.In order to adapt the data of different courses,an adaptive parameter selection method is proposed to search the clustering parameters.(3)A multi-angle learner classification model based on Kmeans is proposed to help teachers provide personalized education services.Based on the feature-weighted algorithm,the course perspective factor is introduced to reflect the impact of the course on the cluster variables.The algorithm correlates the effect of learner's behavior and personal characteristics on the learning result,and could run at various stages of the course,so that making algorithm more universal.(4)The difficulties of predicting learners' certificates are studied in depth,then,an under-sampling stacking model is proposed to overcome the problem of imbalance in dataset and avoid the problem of missing important training samples when usingunder-sampling method.In order to model on a hybrid dataset,course similarity is introduced to share information between different courses.Through the comparison experiment,it is proved that the proposed model has better prediction effect and stronger robustness.