ADMit: Improving NER in automotive domain with domain adversarial training and multi-task learning

Cheoneum Park; Seohyeong Jeong; Juae Kim

首页> 外文期刊>Expert Systems with Application >ADMit: Improving NER in automotive domain with domain adversarial training and multi-task learning

【24h】

ADMit: Improving NER in automotive domain with domain adversarial training and multi-task learning

机译：ADMit: Improving NER in automotive domain with domain adversarial training and multi-task learning

获取原文

获取原文并翻译 | 示例

掌桥外文数据库（机构版） >>

开具论文收录证明 >>

文献代查 >>

页面导航

摘要
著录项
相关主题

摘要

AI virtual assistants are steadily expanding their way into multiple domains such as healthcare, hospitality, and telecommunication. However, such virtual assistants still suffer from a limited level of distinctive characteristics of domain-specific knowledge. In this paper, we improve the Named Entity Recognition (NER) ability by facilitating (ⅰ) the assistants' understanding of both general and domain-specific terms by building new datasets in the automotive domain and using adversarial training and (ⅱ) NER's robustness against spacing errors using multi-task learning. We verify the proposed method in both Korean and English NER datasets and the experimental results demonstrate that the proposed NER model outperforms the baseline model in both general and automotive domains. The proposed method improves automotive NER performance by 1.85 points and 1.13 points in English and Korean, respectively. Furthermore, we validate proposed NER model on the in-house question answering system where queries are taken from real users.

著录项

来源
《Expert Systems with Application》 |2023年第9期|120007.1-120007.12|共12页
作者
Cheoneum Park; Seohyeong Jeong; Juae Kim;
展开▼
作者单位

AIR Lab, Hyundai Motor Group, Seoul, South Korea;

AIR Lab, Hyundai Motor Group, Seoul, South Korea,Department of English linguistics and Language Technology, Hankuk University of Foreign Studies, Seoul, South Korea;

展开▼
收录信息
原文格式 PDF
正文语种英语
中图分类
关键词
Mobility industry; Automotive domain; Named entity recognition; Adversarial training; Multi-task learning;

ADMit: Improving NER in automotive domain with domain adversarial training and multi-task learning

摘要

著录项

相关主题

期刊订阅