Deep Learning Architectures for Named Entity Recognition: A Survey

Abstract

Extracting named entities from natural language text is an important task in natural language processing, with applications in sentiment analysis, information retrieval, and answer selection in question answering. For identifying named entities, many methods have been developed ranging from knowledge-based methods to supervised machine learning methods. In recent years, deep learning models have achieved cutting-edge results in language processing tasks, particularly in Named Entity Recognition (NER). NER aims to locate and categorize proper names in natural language text into predefined classes such as people, organization, location names, etc. In this paper, we provide a comprehensive survey on existing deep learning architectures used in the CoNLL-2003 NER task. We first introduce the basic deep learning models employed in the NER task followed by reviewing the prominent deep learning NER architectures. Furthermore, the present study throws light upon top factors impacting NER performance which includes the design choices of deep-learning-based NER architecture, and the significance of incorporating character-level information, additional lexical features, external training data, etc. while training the NER model.