Automated Essay Scoring in Education

As artificial intelligence advances rapidly, Automated Essay Scoring (AES) technology has made significant strides, effectively addressing the challenges of high labor intensity and slow feedback encountered by teachers during essay grading. In recent years, numerous innovative research findings and methodologies have emerged in this field. We explore in this paper the mainstream studies in AES and categorize them into two groups: single-prompt AES and cross-prompt AES. Single-prompt AES involves scoring scenarios where training and test essays are written for the same prompt, comprising both feature engineering and deep learning-based approaches. On the other hand, cross-prompt AES entails scoring scenarios where training and test essays are written for different prompts, often implemented within the frameworks of contrast learning, transfer learning, or multi-task learning. Drawing from existing research, we analyze the primary issues and challenges in AES and proposes potential directions for future research.