Beyond Accuracy: Detecting and Alleviating Algorithmic Bias in Classifying Educational Forum Messages

Discussion forum is an essential component of online courses. To gain analytical insights from online student discussions on these forums, automatic classifiers were often adopted. However, existing studies mostly stressed the accuracy of a classifier, while the fairness of the classifier remains largely unexplored. Thus, we provided in-depth accuracy and fairness evaluations of popular Machine Learning (ML) models used for educational classification tasks. Further, to empower these models in both accuracy and fairness dimensions, we proposed sampling-based techniques and showed that our approaches can empower ML models in both accuracy and fairness dimensions across different educational predictive tasks.