Research on entity recognition integrating span markers and vocabulary enhancement

Current named entity recognition methods face challenges such as complex entity types and ambiguous boundaries that are difficult to distinguish. To address this, this paper proposes an entity recognition method that integrates span labeling and vocabulary enhancement. First, from the perspective of vocabulary enhancement and pre-trained model integration, vocabulary information is input into the BERT model for feature learning to achieve the purpose of dividing word class boundaries and distinguishing word class attributes. Then, using head and tail annotations, a category pointer network is constructed to solve the entity prediction interruption problem. Finally, the category index is used to decode the entity, comprehensively improving the accuracy of named entity recognition. Experiments demonstrate that the proposed model achieves F1 scores of 96.46% and 72.91% on the Resume and Weibo public datasets, respectively, and outperforms state-of-the-art models, validating the superiority of the proposed approach.