DETG: Dual encoding document level relation extraction based on tree graph

In recent years, graph-based and sequence-based document-level relation extraction models have attracted much attention. However, when performing document-level relation extraction tasks, traditional graph-based models struggle to address the overfitting issue of graph convolutional networks when analyzing document information, and sequence-based models may not effectively capture all document features. To address the above problems, we propose a document level relation extraction model based on dendrogram with dual encoders. The model uses dual encoders to extract document features of various granularities to construct a dendrogram structure. Experimental results on the public dataset DocRED show that the performance of the proposed model is significantly improved compared to other models that solely rely on graph structure or sequence structure.