Knowledge-Enhanced Relation Extraction for Chinese EMRs

The growing demand for the meaningful use of electronic medical records has led to great interest in medical entities and relation extraction technologies. Most existing methods perform relation extraction based on manually labeled documents and rarely consider incorporating knowledge graphs that include rich, valuable structured knowledge, which will cause semantic ambiguities. To address this problem, we propose a knowledge-enhanced relation extraction (KERE) model. First, we extract knowledge information from the knowledge graph to generate a knowledge-guided word embedding. Then, the lexical features are considered supplementary information for semantic understanding. Experiments on real-world datasets show that the KERE model achieves important improvements in a biomedical relation extraction task.