Improving Chinese Named Entity Recognition by Glyph and Part-of-Speech

Chinese characters are a unique type of morpheme-syllabic characters that are integral to the ideographic writing system used in the Chinese language. The most widely used word formation method in Chinese is the formation of compound words from roots according to syntactic relationships, which is basically the same as the syntax of combining words into phrases. However, the appropriate use of glyph information and lexical information in natural language processing tasks is still limited. To solve above problem, this paper proposes a novel approach, named GP-NER, to enhance Chinese character representation by glyph features and inject word information with part-of-speech information. Specifically, GP-NER uses a gate mechanism to enhance the representation of characters by combining glyph features, and then uses part-of-speech information to enhance the contextual representation of text sequences through attention mechanisms. Experimental results demonstrate the effectiveness of the proposed model.