A Deep Learning Based Reasoner for Global Consistency in Named Entity Recognition

Xiaoxiao Yin, Ruifang Liu, Daqi Zheng, Zhengdong Lu

Abstract: Named Entity Recognition (NER) is a basic task of Natural Language Processing (NLP), it’s a challenging task in a variety of special applications. This paper aims to solve the global consistency of NER, and to improve the performance. Inspired by human reading process, we propose a NE-Reasoner model, which combine deep neural networks and memory artificial neural network to identify named entities with global consistency. The advantages of the model are: (1) The multi-layer deep architecture, allowing it to bootstrap the recognized entity set from coarse to fine. (2) The candidate pool memory mechanism, allowing it to exchange identified entity information between layers. (3) The reasoner, combing encoder-decoder and cached information to infer to get global entities. The experimental results show that the NE-Reasoner can identity ambiguous words and named entities that rarely or never met before.