An End-To-End NER Model with Explicit Boundary and Type Information

Abstract

Named Entity Recognition (NER) is a fundamental task in Natural Language Processing (NLP). The purpose is to locate and classify named entities with specific meanings, including the person and location. However, there are two problems with the existing methods: Firstly, the model cannot significantly distinguish boundaries and types of entities, especially if there are many types of entities. Secondly, the labels do not provide information on the differences in the boundaries and types of model entities. Therefore, the Explicit Boundary and Type Information (EBTI) models are used to address both issues to improve NER performance. Specifically, the boundary detection module uses for pure boundary recognition, and the type module uses to determine entity type information and then combine them to improve the performance of NER. Experiments are conducted on the widely used dataset CoNLL2003 to demonstrate the advantages of our method, with results comparable to the current state-of-the-art (SOTA).