Enhancing Essay Scoring: An Analytical and Holistic Approach With Few-Shot Transformer-Based Models

Abstract

In the field of automated essay scoring (AES), the task of evaluating written compositions has been a persistent challenge. Despite the impressive capabilities of generalized transformer models in various natural language processing (NLP) domains, their application to essay scoring has often fallen short of expectations. In response to this ongoing challenge, this research delves into the intricate nuances of holistic and analytical essay assessment. This work presents an innovative approach centered on Few-Shot transformer-based models, capitalizing on the strengths of pretrained language models while enabling fine-tuning with limited essay-specific data, often called ‘Few-Shot.’ The outcomes of this study are highly promising, with significant improvements in essay scoring accuracy that surpass the performance benchmarks established by conventional methods. The proposed methodology demonstrates remarkable enhancements in the Quadratic Weighted Kappa (QWK) score, indicating its potential. This represents a significant stride towards automating sophisticated essay evaluation, addressing a long-standing issue in the field.

Keywords

• AES
• NLP
• transfer learning
• BERT
• few-shot learning
• holistic scoring