An Interactive Short Answer Grading System Based on Active Learning Models

Grading automation can improve learning experience with quick around-the-clock feedback and superior grading consistency. Obtaining annotated data for training short answer grading models is costly. Active learning has been proven an effective approach to build accurate models with few annotated data. This paper presents an active learning approach of short answer grading that comprises of a few novelties. The first is a specialized active learning formulation adapted to short answer grading principles. The second is a proposal to exploit human expertise in fine-tuning several active learning model parameters for adaptation to the specifics of each grading task. The third is an interactive short answer grading system that is designed for building better quality grading model by informing users with data visualizations. The prototype presented in the paper should provide a useful conceptual demonstration for real-life deployment of active learning for short answer grading and further research in an enhanced interactive form of active learning.