Attention-Based Bidirectional Long Short-Term Memory Neural Network for Short Answer Scoring

The automatic short answer scoring by using computational approaches has been considered the best way to release the workload of human answer raters. In this paper, we designed a novel neural network architecture which is attention-based bidirectional long short-term memory to implement the task of automatic short answer scoring. We evaluate our approach on the Kaggle Short Answer dataset (ASAP-SAS). Our experiment results indicate that our model can scoring short answers more accurately in terms of the quality of the results. Meanwhile, our experiment results demonstrate that our model is more effective and efficient than other baseline methods in most cases.