Joint Entity-Relation Extraction Model for Epidemiological Reports Based on SMART Adversarial Training

Epidemiological investigation work requires the drafting of complex epidemiological reports, the information within which is crucial for tracing the origin of outbreaks and for preventive measures against future epidemics. Currently, each segment of text in these reports contains a significant number of triplets, most of which are overlapping. Traditional entity-relation extraction algorithms struggle to handle these overlapping triplets. In response to this situation, this paper introduces a joint entity-relation extraction model based on SMART adversarial training and the CasRel network. Through comparative experiments conducted on both the dataset constructed for this paper and publicly available datasets, the feasibility and effectiveness of the proposed model have been verified.