Bi-directional Joint Embedding of Encyclopedic Knowledge and Original Text for Chinese Medical Named Entity Recognition

Abstract

Named entity recognition is a basic task in natural language processing and can be used for building knowledge graph. Named entities generally refer to the entities in the text that have a specific meaning or a strong reference. In medical area, a great deal of medical information exists in the form of electronic text and we can acquire valuable part by this method. In this article, we take extra knowledge information of medical encyclopedia into account and we associate the original text in the named entity recognition task with its encyclopedic knowledge to enhance the ability of entity recognition through the establishment of the connection and interaction of the joint-network. Based on this, (1) the attention distribution based on medical encyclopedia knowledge was generated with entity embedding and context information in the sentence, which can be directly used to generate encyclopedia knowledge subnet; (2) the bi-directional joint embedding model integrates knowledge subnet and text subnet into one network so that the connection between them are considered. We conduct experiments on tow electronic medical record datasets proposed by CCKS. The experimental result shows that our proposed method has achieved a better result.