A Transferable Framework: Classification and Visualization of MOOC Discussion Threads

Lin Feng, Guochao Liu, Sen Luo, Shenglan Liu

Abstract: Analysis of Massive Open Online Course (MOOC) forums data often use natural language processing (NLP) technology to extract keywords from discussions content as features. However, discussions in different course forums vary significantly, so the analyzed results obtained on these specific forums are not easily applied to other irrelevant forums. Besides, a lot of discussion threads are not related to the course in MOOCs forums. To address above problems, we analyze about 100,000 discussion threads from the forums of 60 MOOCs offered by Coursera, and design many features related to user interaction ways in different subforums. This work proposes a transferable framework to classify MOOC discussion threads using these features. The classification framework is not sensitive to the subjects and the forum discussions, so the classification model can be used directly by other course forums without being trained again. Experiments show that the average classification performance with Area Under ROC Curve (AUC) is 0.8. This work also gives the methods to remove noisy discussion threads and visulize the interactive characteristic of MOOC forum threads using dimensionality reduction technology.