Abstract: NER(Named Entity Recognition) is a fundamental task in natural language processing,with extensive applications in information extraction,question answering systems,and knowledge graph construction.However,existing approaches still struggle with inadequate multi-scale feature utilization and inaccurate boundary identification when processing nested entities and ambiguous entity boundaries in Chinese text.To tackle these challenges,this paper proposes a Chinese NER model incorporating an AAM(Adaptive Attention Mechanism) and a BEM(Boundary Enhancement Module),specifically designed to handle the absence of explicit word delimiters and complex semantic structures in Chinese.The AAM dynamically integrates local and global contextual features to enhance the modeling of intricate Chinese semantic patterns,while the BEM employs depthwise convolution to strengthen boundary perception,effectively reducing recognition errors caused by nested entities and ambiguous spans.Experimental results demonstrate that the proposed model achieves F1 scores of 94.39% and 83.72% on the nested Chinese datasets ACE2005-Chinese and Cnerta,and 77.75%,84.88%,and 96.36% on the flat Chinese datasets Weibo,Ontonotes,and Resume,consistently surpassing existing mainstream Chinese NER methods and validating its effectiveness and generalization capability across diverse Chinese text scenarios.

Key words: Chinese named entity recognition, Adaptive attention, Boundary enhancement, Nested entity, Multi-scale features