BDCP: An Improved Nested Named Entity Recognition Model Based on LSTM
Ling Zhou, Tianjiao Luan, Na Yao, Jianming Li, Jie Ding · 2021
In construction of knowledge graphs (KGs), named entity recognition (NER) is a sub-task to identify the boundaries of entities with special meaning and predict their categories in texts. The instance that an entity contains one or more entities is called nested entity. The current work on NER usually ignores the nested NER. We propose an improved model named Boundary Detection and Category Prediction (BDCP) for Nested NER. Our model uses LSTM to extract the contextual features of the sequence, and uses boundary detection unit to mark the boundaries of entities in the text. By introducing boundary detection unit, our model extracts the boundaries of entities and restrict the number of candidate entities. We also design a layer-by-layer decoding module by boundary detection unit for Nested NER. Experiments on Nested NER datasets named GENIA [1] demonstrate the effectiveness of our model on nested NER.