Comparative Study of Two Different Mooc Forums Posts Classifiers: Analysis and Generalizability Issues

Abstract

Massive Open Online Courses (MOOCs) offer a wide range of opportunities for learning. Their growing popularity has resulted in a large amount of data being available for learning analytics purposes. A major problem of MOOCs is the overwhelming number of posts in their discussion forums. The forum is a key part of the learning process within a MOOC, so this information overload affects negatively the participants’ learning experience. Automatic classification of the posts can help searching of relevant information for both the learners and teaching assistants. In this study, we address this problem by building two multiclass classification models, using natural language processing techniques, that classify the posts according to a three-category coding scheme. Each model was created with data derived from a MOOC of different subject matter. The main goal was to evaluate each model’s accuracy along with its generalizability to courses of different subject matter. This study contributes to the line of research for automatic classification of forum discussions, ultimately aiming at the development of tools that may assist participants while searching in the forum. Furthermore it provides insights on the main issues that inhibit generalization of classifiers created for a specific subject matter and investigate how their linguistic features relate to this inhibition.