MOOC Forum Discussion Threads Classification Research Based On Machine Learning

Along with the development of Massive open online courses(MOOC),online education has become one of the most popular educational models nowadays,how to improve the teaching quality of online education has also become a popular research direction in data mining.MOOC forum is the only platform for students to communicate with teachers and assistants in MOOC courses,and and it is an important factor directly related to the quality of the whole course.A reasonable and accurate classification of the thread of the forum can help students to communicate and ask questions better,and to solve the difficulties encountered in the study more efficiently.Traditional analysis of MOOC forums data often use natural language processing or text analysis technology,first extract the forums text keywords,then utilize this keywords to extract text features,build a classifier to distinguish the forms discussion threads.However,these methods can't transform the research findings from specific forums to other forums,because the difference of course forums content is very huge,and the languages for forum users use is variety.To address above problems,this paper proposes a thread classification framework of MOOC forum based on user behavior features.This paper firstly analyzes the data of 60 courses of Coursera,which is the largest MOOC platform in the world,also collect and analyzes 3 most popular courses of Coursera,through these work we can get that user behavior feature can also be used to distinguish different kinds of MOOC forum threads very well.Based on the structure,potential social network,popularity,and quality of forum discussion threads,we propose 23 types of user behavior features.User behavior features are completely irrelevant to text information,so it can be extract from all kinds of MOOC forums and the classification model can be used on other MOOC forums directly.Experiments show a classification performance with ROC-AUC is 0.8,it is 12% higher than other results.The design and selection of user behavior features are very important for the accuracy of the thread classification model,but the design and selection of the features usually require a strong priori knowledge and need to consume a lot of human cost.In order to solve this problem,this paper proposes that the gradient boosting decision tree can be used to select and combine the original user behavior features,which can automatically discover more discriminant features and combination features,thus reducing the difficulty of manual design of user behavior.Using gradient booting decision tree to encode the features,it can bring more than 5% growth on the basis of the best classification results.