AESPrompt: Self-supervised Constraints for Automated Essay Scoring with Prompt Tuning
Qiuyu Tao, Jiang Zhong, Rongzhen Li · Proceedings/Proceedings of the ... International Conference on Software Engineering and Knowledge Engineering · 2022
Automated essay scoring(AES) aims to automatically assign scores to essays based on the quality of writing.Previous approaches have made many attempts with pre-trained BERT for essay scoring and achieved the state-of-the-art.However, these approaches mainly rely on the high computation cost and ignore the high similarity between text representations.In this paper, we propose a lightweight prompt-tuning framework, AESPrompt, to capture the significant semantic features of the text efficiently.We construct one continuous prompt for each layer of the frozen language model to help the language model understand the essay scoring task.Specially, we design taskrelated self-supervised constraints to capture discourse structure in terms of coherence and cohesion further to enhance the generalization and discourse awareness of the prompt.Experimental results on the public dataset ASAP illustrate that our approach performs competitively in the full data settings and outperforms in one-shot data settings significantly compared with fine-tuning BERT.