SDTCNs: A Symmetric Double Temporal Convolutional Network for Chinese NER

Abstract

Listen

Translate article

Chinese NER is a basic task of Chinese natural language processing. Most current models for Chinese NER can be roughly divided into two categories: character-based models and word-based models. Character-based models cannot effectively utilize the inherent information of a word. Word-based models cannot effectively disambiguate words under different word segmentation norms. In this paper, we propose a symmetric double temporal convolutional network for Chinese NER: SDTCNs. SDTCNs is built on the BERT model and is composed of two symmetric temporal convolutional networks, where one is used to learn the location features of named entities and the other is used to learn the class features of named entities. Finally, a fusion algorithm proposed in this paper is used to fuse location features and class features to obtain the final named entity. Experiments on various datasets show that SDTCNs outperforms multiple state-of-the-art models for Chinese NER, achieving the best results.