GDPNet: Refining Latent Multi-View Graph for Relation Extraction

Relation Extraction (RE) is to predict the relation type of two entities that are mentioned in a piece text, e.g., sentence or dialogue. When given text long, it challenging identify indicative words for prediction. Recent advances on RE task from BERT-based sequence modeling and graph-based relationships among tokens sequence. In this paper, we propose construct latent multi-view graph capture various possible tokens. We then refine select important Finally, representation refined concatenated extraction. Specifically, our proposed GDPNet (Gaussian Dynamic Time Warping Pooling Net), utilize Gaussian Graph Generator (GGG) generate edges graph. The by (DTWPool). On DialogRE TACRED, show achieves best performance dialogue-level RE, comparable with state-of-the-arts sentence-level RE.