LE-NER: A Chinese NER Model Based on Lexical Enhancement

Dong Li, Shumei Du, Peixuan Li, Baoyan Song, Zhicong Liu, Yue Kou

Abstract: This paper proposes a novel Chinese Named Entity Recognition (NER) Model Based on Lexical Enhancement (termed LE-NER). Firstly, the model integrates character and lexical information at the character representation stage, significantly enhancing feature representation and effectively reducing segmentation errors due to ambiguous entity boundaries. Secondly, the LE-NER model employs a residual stacked BiLSTM network with inter-layer attention fusion to model input sequences, dynamically assigning varying weights to features from different layers, thereby accurately extracting the most relevant feature information. This optimization strategy notably enhances the effectiveness of text entity recognition, improving accuracy and efficiency in processing complex textual data. Finally, experiments conducted on publicly available datasets, including OntoNotes, Weibo, Resume, and MSRA, validate the feasibility and effectiveness of the proposed LE-NER model.