结合对抗训练的双通道特征融合命名实体识别OA

Existing named entity recognition(NER)models often lack focus on local information and face challenges in handling long-range dependencies and complex sequence data.To address these is-sues,a dual-channel feature fusion NER model incorporating adversarial training was proposed.First,features were extracted using a pretrained model,with adversarial training enhancing robustness and gen-eralization.Then,a dual-channel module,comprising multi-head attention and time steps,was intro-duced after a bidirectional long short-term memory network to capture both global and local information,improving the model's ability to manage long-range dependencies and complex sequences.Finally,GlobalPointer and rotary position encoding were employed to identify the head and tail information of enti-ties,reducing computational costs while further enhancing the model's ability to capture global informa-tion.Experimental results showed that the proposed model achieved improvements in F1 scores across five public datasets,validating its effectiveness in capturing both global and local information and han-dling complex sequence data.