Research on Named Entity Recognition Methods for Government-Related Queries

This paper proposes a named entity recognition method based on the characteristics of Chinese government domain policy documents to improve the accuracy of named entity recognition (NER) for government-related queries, and develops a NER model specifically for government named entity recognition. To address challenges such as the lack of publicly available data sets, the complexity of Chinese text structures, and the uneven distribution of entities. Model design includes strategies for independently building and annotating government-relevant data sets to mitigate data scarcity issues. According to the characteristics of Chinese text, a sequence labeling strategy is designed to improve the handling of entity diversity and context dependence. In addition, the model is optimized to improve the recognition performance of low-frequency entities. The results show that compared with other models, this model can accurately identify various complex entities in the government field and show certain performance advantages.