Biomedical Causal Relation Extraction Incorporated with External Knowledge

Biomedical causal relation extraction is an important task. It aims to analyze biomedical texts and extract structured information such as named entities, semantic relations and function type. In recent years, some related works have largely improved the performance of biomedical causal relation extraction. However, they only focus on contextual information and ignore external knowledge. In view of this, we introduce entity information from external knowledge base as a prompt to enrich the input text, and propose a causal relation extraction framework JNT_KB incorporating entity information to support the underlying understanding for causal relation extraction. Experimental results show that JNT_KB consistently outperforms state-of-the-art extraction models, and the final extraction performance F1 score in Stage 2 is as high as 61.0%.