VisPhone: Chinese named entity recognition model enhanced by visual and phonetic features

Baohua Zhang, Jiahao Cai, Huaping Zhang, Jianyun Shang

Abstract: Highlights•Proposal for a NER model combining phonetic and visual features of Chinese characters.•We believe Chinese pronunciation has cultural meaning and have proof.•We tested various combinations of phonetic features and found the best one.•Cross transformer effectively combines phonetic, visual info with text embedding.•The model achieves excellent performance on four well-known benchmark datasets.