MGCN: Medical Relation Extraction Based on GCN

keywords: Relation extraction, co-occurrence graph, attention mechanism, open-world assumption, graph convolutional network
With the progress of society and the improvement of living standards, people pay more and more attention to personal health, and WITMED (Wise Information Technology of med) has occupied an important position. The relationship prediction work in the medical field has high requirements on the interpretability of the method, but the relationship between medical entities is complex, and the existing methods are difficult to meet the requirements. This paper proposes a novel medical information relation extraction method MGCN, which combines contextual information to provide global interpretability for relation prediction of medical entities. The method uses Co-occurrence Graph and Graph Convolutional Network to build up a network of relations between entities, uses the Open-world Assumption to construct potential relations between associated entities, and goes through the Knowledge-aware Attention mechanism to give relation prediction for the entity pair of interest. Experiments were conducted on a public medical dataset CTF, MGCN achieved the score of 0.831, demonstrating its effectiveness in medical relation extraction.
mathematics subject classification 2000: 68U15
reference: Vol. 42, 2023, No. 2, pp. 411–435