Char-word fusion method for Chinese NER

Letian Lu; Hao Liu

doi:10.1117/12.2624996

6 December 2021 Char-word fusion method for Chinese NER

Letian Lu, Hao Liu

Proceedings Volume 12085, International Conference on Green Communication, Network, and Internet of Things (GCNIoT 2021); 120850P (2021) https://doi.org/10.1117/12.2624996
Event: 2021 International Conference on Green Communication, Network, and Internet of Things, 2021, Kunming, China

Abstract

The task of Chinese named entity recognition (CNER) is closely related to Chinese word segmentation, because most Chinese entities are composed of words. The CNER model with words as the minimum input unit is called word based CNER model. However, wrong word segmentation will lead to wrong entity recognition results, so the CNER model with single word as input (char based) has gradually become the focus of research. Recently, many efforts have begun to integrate external word information to improve the performance of word based CNER model. And softlexicon are excellent word information fusion methods. When using softlexicon method for word fusion, too much matching word set information will be introduced, resulting in additional "noise" interference, resulting in entity recognition errors. Aiming at the problems of softlexicon method, this paper proposes a word information Selected-Softword-Concat method (SSC), which effectively solves the problem of "noise" in softlexicon method by introducing word segmentation information. The experimental results on four general CNER data sets show that compared with softlexicon method, the F1 value of the model constructed by SSC method is increased by 0.94, 0.92, 0.78 and 0.25 respectively, which effectively solves the "noise" problem of softlexicon.

Citation Download Citation

Letian Lu and Hao Liu "Char-word fusion method for Chinese NER", Proc. SPIE 12085, International Conference on Green Communication, Network, and Internet of Things (GCNIoT 2021), 120850P (6 December 2021); https://doi.org/10.1117/12.2624996

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available