关键词: 命名实体识别, 深度学习, 分布式输入表示, 上下文编码, 标签解码器

Abstract: To address the problems of knowledge extraction and representation difficulties caused by the coexistence of multimodal heterogeneous data in advanced manufacturing industry,the methods of Named Entity Recognition (NER) methods  become the research hot spot.The commonly used methods of NER in industry was summarized,which had introduced the traditional methods of NER firstly and then expounded the methods based on deep learning.The common methods were classified and analyzed from  three stages of distributed input representation,context encoder and tag decoder.The various methods of distributed input representation were compared,and their advantages and disadvantages were pointed out.The context encoder models from long-distance dependence capture,local context information,parallelism,information loss degree and transportability were also compared,and the characteristics of each model were given.Finally,the challenges to be addressed and research directions in the future were illustrated.

Key words: named entity recognition, deep learning, distributed input representation, context encoder, tag decoder