Named entity recognition for Chinese marine text with knowledge-based self-attention

Chinese named entity recognition has been widely used in many fields, such as species recognition in marine information, and so on. Compared with the standard named entity recognition (NER), the performance of the Chinese marine named entity recognition is low, which is mainly limited by the normative nature of the text and the scale of the tagged corpus. In recent years, the research of named entity recognition primarily focuses on the small scale of the tagged corpus. It tends to use external knowledge or use joint training to improve final recognition performance. However, there is little research on the problem that the accuracy of NER will be suboptimal if the model is trained inadequately. To this end, in order to improve the accuracy of naming entity recognition and recognize more entities in corpus, this paper proposes a named entity recognition method that combines knowledge graph embedding with a self-attention mechanism. The entity embeddings of the marine knowledge graph (KG) are empolyed for the hidden units of NER model with attention mechanism in an end-to-end way. Therefore, the model can get additional auxiliary information to improve performance. Lastly, we conduct extensive experiments on marine corpus and other public datasets. The experimental results verify the effectiveness of our proposed method.