Semi-Automatic Labeling of Online Course Discussion Posts

Dharamjit Parmar, M. Ali Akber Dewan, Dunwei Wen, Fuhua Oscar Lin

Abstract: Automatic text labeling is essential in diverse text analysis applications, which require high precision. This paper presents a semi-automatic approach for text labeling by combining deep learning methods with human interventions. We employ bidirectional long short-term memory (Bi-LSTM) and convolutional neural network (CNN) to generate initial labels of online course discussion posts, which we then refine through a structured human-in-the-loop feedback mechanism. This semi-automatic and iterative process of labeling the forum posts reduces time for the labeling task and enhances the reliability of the training data to be used in real application of text classification. We evaluated our framework on a MOOC course dataset, demonstrating significant improvement in model performance. The results underscore the potential of integrating human expertise to complement and augment machine learning in automating the labeling tasks, paving the way for more reliable and robust applications of text analysis, especially in education.