Document-Level Relation Extraction with Cross-sentence Reasoning Graph

Relation extraction (RE) has recently moved from the sentence-level to document-level, which requires aggregating document information and using entities mentions for reasoning. Existing works put entity nodes mention with similar representations in a document-level graph, whose complex edges may incur redundant information. Furthermore, existing studies only focus on entity-level reasoning paths without considering global interactions among cross-sentence. To these ends, we propose novel RE model GRaph Aggregation Cross-sentence Reasoning network (GRACR). Specifically, simplified graph is constructed semantic of all sentences document, an designed explore relations long-distance cross-sentence pairs. Experimental results show that GRACR achieves excellent performance two public datasets RE. It especially effective extracting potential Our code available at https://github.com/UESTC-LHF/GRACR .