지도식 미세조정(SFT)을 이용한 철자오류 교정. (Korean)

Spelling Error Correction using Supervised Fine-Tuning(SFT). (English)

This paper focuses on the important topic of spelling error correction in the field of natural language processing. The research experimentally explores spelling error correction methods using the Generative Pre-trained Transformer (GPT) model. Specifically, it aims to improve the performance of spelling error correction by employing supervised fine-tuning methodologies applied in the GPT-1 and InstructGPT models. The supervised fine-tuning in this paper utilizes direct human feedback to improve the model's performance. It also introduces various learning prompt strategies to further enhance the performance of the spelling error correction model. These strategies, which utilize bidirectional context information for training, showed significant improvement in F1 scores compared to the baseline GPT-2 model, achieving higher performance than previous research results. This study demonstrates the effectiveness of employing GPT models for spelling error correction, as well as the methodologies of supervised fine-tuning and learning prompt strategies. These findings present the potential to advance research in this field. [ABSTRACT FROM AUTHOR]

Copyright of Journal of the Korea Institute of Information & Communication Engineering is the property of Korea Institute of Information & Communication Engineering and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)