FGN: Fusion Glyph Network for Chinese Named Entity Recognition

Abstract

Chinese NER is a challenging task. As pictographs, Chinese characters contain latent glyph information, which is often overlooked. In this paper, we propose the FGN 1 , Fusion Glyph Network for Chinese NER. Except for adding glyph information, this method may also add extra interactive information with the fusion mechanism. The major innovations of FGN include: (1) a novel CNN structure called CGS-CNN is proposed to capture both glyph information and interactive information between glyphs from neighboring characters. (2) we provide a method with sliding window and Slice-Attention to fuse the BERT representation and glyph representation for a character, which may capture potential interactive knowledge between context and glyph. Experiments are conducted on four NER datasets, showing that FGN with LSTM-CRF as tagger achieves new stateof-the-arts performance for Chinese NER. Further, more experiments are conducted to investigate the influences of various components and settings in FGN.

A framework and prototype system in support of workflow collaboration and knowledge mining for manufacturing value chains

“…Furthermore, due to the development of artificial intelligence and knowledge graphs, a lot of research has been done to discuss how to extract knowledge from various forms of data such as plain text, images, and other unstructured data [26–29]. In this paper, knowledge refers to the form of $$\left({e}_{1},r,{e}_{2}\right)$$ triples, where $${e}_{1}$$ and $${e}_{2}$$ are two named entities, and $$r$$ is a directed relation between $${e}_{1}$$ and $${e}_{2}$$.…”

“…For example, BERT [31] is the first PLM that uses deep bidirectional transformers to learn representations from unlabelled text and perform significantly improved for a wide range of tasks. Various studies have exploited pre-trained LMs to extract knowledge from unstructured data in different domains, such as entity recognition [26][27][28], relation extraction [29,34], and joint extraction [35][36][37]. FLAT [26] uses a lattice structure and an elaborate transformer to perform Chinese named entity recognition (NER).…”

AIP: A Named Entity Recognition Method Combining Glyphs and Sounds

“…FGN was proposed by Xuan et al On the one hand, a new CNN structure was proposed, called CGS-CNN. On the other hand, a method with a sliding window and Slice-Attention to fuse the BERT representation and glyph representation for a character was provided [28]. Recently, some scholars paid attention to not only the glyph features but also the pinyin features of Chinese characters.…”

Robust Self-Augmentation for Named Entity Recognition with Meta Reweighting

“…In recent years, research on NER has concentrated on either enriching input text representations (Zhang and Yang, 2018;Nie et al, 2020b;Ma et al, 2020) or refining model architectures with various external knowledge (Zhang and Yang, 2018;Ye and Ling, 2018;Li et al, 2020a;Xuan et al, 2020;Li et al, 2020b;Shen et al, 2021). Particularly, NER model, with the aid of large pretrained language models (Peters et al, 2018;Devlin et al, 2019;, has achieved impressive performance gains.…”