Research on Named Entity Recognition Based on Bidirectional Pointer Network and Label Knowledge Enhancement.

Named Entity Recognition (NER) is one of the fundamental tasks in natural language processing (NLP) and serves as a foundation for many downstream tasks. Currently, most span-based methods utilize sequence labeling to detect entity boundaries. This method may cause too many invalid entities and the number of entity fragments explosion when combining entity boundaries to generate entity fragments. To solve these problems, we propose a model based on bidirectional pointer network and label knowledge enhancement. We used pointer network mechanism to detect entity boundary, pointing entity beginning boundary to entity ending boundary, and non-entity to sentinel word. This processing method can effectively alleviate the problem of excessive invalid entity and explosion of entity fragment quantity. In addition, we also fully integrate the label knowledge into the model to improve the effect of the model. We conducted extensive experiments on the ACE2004 dataset and achieved better results than most existing named entity recognition models.