Syntactic methods for document-level relation extraction use features derived from dependency parse trees, often alongside coreference annotations and discourse relations, to connect information across sentences and identify relations expressed in complicated document contexts.

Columbia University

Google

- Syntactic methods
- Multi-hop paths between entities
- Graph-structured neural networks

Methods for handling more complicated context in RE

Inter-sentence entity graphs can use multi-hop paths between entities for inferring the correct relations. 


Multi-hop paths between entities for RE from complicated context


Graph-structured neural networks build a graph over the sentences or entities in a document and use graph neural network layers to model cross-sentence dependencies for relation extraction. Propagating information across this graph structure gives the model memory and reasoning abilities that single-sentence models lack.

Learn Before

Related