Improving Automated Essay Scoring by Prompt Prediction and Matching

Area of Science:

• Natural Language Processing
• Educational Technology
• Artificial Intelligence

Background:

• Automated essay scoring (AES) is a key application of NLP in education.
• Pre-trained models are increasingly used for AES, but prompt feature extraction needs improvement.
• Current methods lack sufficient focus on optimizing prompt features for fine-tuning.

Purpose of the Study:

• To develop an effective prompt feature fusion method for AES.
• To enhance feature extraction from pre-trained encoders for better essay scoring.
• To investigate the impact of multi-task learning on AES performance.

Main Methods:

• A novel prompt feature fusion method was created for fine-tuning.
• Multi-task learning was employed with two auxiliary tasks: prompt prediction and prompt matching.
• The NEZHA pre-trained encoder was utilized and evaluated.

Main Results:

• Both auxiliary tasks individually improved model performance in AES.
• The combination of auxiliary tasks with the NEZHA encoder yielded the best results.
• Quadratic Weighted Kappa improved by 2.5% and Pearson's Correlation Coefficient by 2% on average.

Conclusions:

• Multi-task learning and prompt feature fusion are effective strategies for enhancing AES.
• The proposed method significantly boosts the performance of pre-trained models in educational applications.
• This research offers a promising direction for more accurate automated essay evaluation.