ESN-NER

Abstract

Chinese named entity recognition (NER) is more difficult than it in English because of the lack of nature delimiters. First, Chinese NER requires word segmentation, but word-based segmentation will generate errors due to the different granularity of the word segmentation tools. Second, most NER models heavily rely on local linguistic features, but the scope of influence provided by local linguistic features is limited, so sometimes the model will give different results to the same entity in different sentences. To address the above problems, we propose the Entity Storage Network Model called ESN Model for Chinese NER, which is a character-based model to avoid word segmentation errors. Specifically, we design an entity storage layer in this model to extract and store the entity information as a local linguistic feature, and design a position feature which is generated by four flags to enhance the learning of boundary. Then we incorporate the attention mechanism to extend the scope of the local linguistic features. The experimental results on two real-world datasets demonstrate that our model outperforms the state-of-the-art models in Chinese NER task.