Entity Structure Within and Throughout: Modeling Mention Dependencies for Document-Level Relation Extraction

Entities, as the essential elements in relation extraction tasks, exhibit certain structure. In this work, we formulate such entity structure distinctive dependencies between mention pairs. We then propose SSAN, which incorporates these structural within standard self-attention mechanism and throughout overall encoding stage. Specifically, design two alternative transformation modules inside each building block to produce attentive biases so adaptively regularize its attention flow. Our experiments demonstrate usefulness of proposed effectiveness SSAN. It significantly outperforms competitive baselines, achieving new state-of-the-art results on three popular document-level datasets. further provide ablation visualization show how guides model for better extraction. code is publicly available.