Knowledge-Enhanced Relation Extraction in Chinese EMRs

Electronic Medical Records (EMRs) is one of the important data sources of clinical information. Relation extraction is a key step to extract rich medical knowledge from EMRs, which has been studied by many scholars. However, there are some problems in EMRs corpus, such as entity nesting and relation overlapping, which make it difficult to achieve ideal results in EMRs relation extraction task. Previous studies rarely considered the fusion of knowledge graph containing rich and valuable structured knowledge, which leads to semantic ambiguity and other issues. Aiming at the above problems, Relation Extraction model based on Knowledge Graph and Chinese character Radical information(RE-KGR) model is proposed in this paper to study the relation extraction of EMRs in diabetic patients. Firstly, knowledge information is extracted from knowledge graph and embedded by GCN. At the same time, the corresponding radical features of Chinese characters are fused to enhance the semantic information of the input text. Compared with other baseline models, the DEMRC and DiaKG experiments of EMRs datasets of diabetic patients were improved by 1.32% and 2.19%.