Medical Relation Extraction Based On Feature Fusion Forest

Medical relation extraction has received extensive research and application. However, due to the high information density of biomedical medical texts, the extraction performance of relation extraction models in the medical field is not satisfactory. This paper proposes a two-dimensional feature fusion method. We define semantic and syntactic information, as well as word-level and sentence-level information, as features on two dimensions, and further use two attention mechanisms to fuse the features on the two dimensions to improve the performance of relation extraction models. We conducted experiments on three datasets and demonstrated that our model performs better than current relation extraction models in biomedical medical texts.