Extracting Urgent Questions from MOOC Discussions: A BERT-Based Multi-output Classification Approach

Abstract

Online discussion forums are widely used by students to ask and answer questions related to their learning topics. However, not all questions posted by students receive timely and appropriate feedback from instructors, which can affect the quality and effectiveness of the online learning experience. Therefore, it is important to automatically identify and prioritize student questions from online discussion forums, so that instructors can provide better support and guidance to the students. In this paper, we propose a novel hybrid convolutional neural network (CNN) + bidirectional gated recurrent unit (Bi-GRU) multi-output classification model, which can perform this task with high accuracy and efficiency. Our model consists of two outputs: the first one classifies whether the post is a question or not, and the second one classifies whether the classified question is urgent or not urgent. Our model leverages the advantages of both CNN and Bi-GRU layers to capture both local and global features of the input data, as well as the Bidirectional Encoder Representations from Transformers (BERT) model to provide rich and contextualized word embeddings. The model achieves an F1-weighted score of 94.8% when classifying whether the posts are questions or not, and obtains an 88.5% F1-weighted score while classifying the question into urgent and non-urgent. Distinguishing and classifying urgent student questions with high accuracy and coverage can help instructors provide timely and appropriate feedback, a key factor in reducing dropout rates and improving completion rates.