Nested Entity Recognition Fusing Span Relative Position and Region Information

At present, span-based entity recognition methods are mainly used to accurately identify the span (entity) boundary for recognition, in which relative position information of and words region routinely ignored. This can be improve performance. Therefore, a nested model, integrates within span, is proposed. The representation first obtained with triaffine attention. Then, word region, as well previous representation, fused obtain new label-level another Finally, task carried out by cooperative biaffine mechanism. Experiments were conducted on some public datasets, including ACE2004, ACE2005 GENIA. results show that F1-scores achieved using proposed method 87.66%, 86.86% 80.90% GENIA, respectively. These experiments state-of-the-art (SOTA) results. Moreover, model has fewer parameters needs resources lower time complexity than existing mechanism model.