Robust Automated Essay Scoring by Using Attentive Capsule

The computer-assisted essay scoring systems can provide accurate scores in a short time, relieving the stress of teachers assessing student essays, which has attracted widespread attention from academia and industry fields. Considering the existing advanced NLP techniques, we propose a robust end-to-end AES model by leveraging the capsule network framework with multiple losses. The proposed model has testified with baseline models on the public Automated student Assessment Prize (ASAP) dataset. The results demonstrate that the proposed model achieved the stateof-the-art result of the ASAP dataset. The experiment results in the ASAP corpus proved that the proposed model can further improve the accuracy of the PLM model for the AES task.