KCC 2021
Current Result Document : 33,480 / 33,480
| 한글제목(Korean Title) |
BERT Pre-trained Models for Data Augmentation in Twitter Medical Named-Entity Recognition |
| 영문제목(English Title) |
BERT Pre-trained Models for Data Augmentation in Twitter Medical Named-Entity Recognition |
| 저자(Author) |
Kokoy Siti Komariah
Bong-Kee Sin
|
| 원문수록처(Citation) |
VOL 48 NO. 01 PP. 0870 ~ 0872 (2021. 06) |
한글내용
(Korean Abstract) |
|
영문내용
(English Abstract) |
Data augmentation is a technique often employed to increase the size of training data that is the same as the original data synthetically. However, in Named Entity Recognition tasks that makes predictions at the token level, it is difficult to augment a set of words without changing the existing label and context of the sentence. In this paper, we take BERT to generate a new sentence by predicting the masked words and replace them according to both the context and its label. Experiments on six different BERT pre-trained models show that data augmentation using a deep bidirectional language model can generate more data with a relevant context in a short text such as tweets and improve the classifier's performance. |
| 키워드(Keyword) |
data augmentationmedical named-entit
bert pre-trained modeltwitter dat
information extractio
|
| 파일첨부 |
PDF 다운로드
|
