Document-Level Relation Extraction with Path Reasoning

Document-level relation extraction (DocRE) aims to extract relations among entities across multiple sentences within a document by using reasoning skills (i.e., pattern recognition, logical reasoning, coreference reasoning, etc.) related to the reasoning paths between two entities. However, most of...

Full description

Saved in:
Bibliographic Details
Published in:ACM transactions on Asian and low-resource language information processing 2023-03, Vol.22 (4), p.1-14, Article 104
Main Authors: Xu, Wang, Chen, Kehai, Zhao, Tiejun
Format: Article
Language:English
Subjects:
Computing methodologies
Information extraction
Citations: Items that this one cites
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Document-level relation extraction (DocRE) aims to extract relations among entities across multiple sentences within a document by using reasoning skills (i.e., pattern recognition, logical reasoning, coreference reasoning, etc.) related to the reasoning paths between two entities. However, most of the advanced DocRE models only attend to the feature representations of two entities to determine their relation, and do not consider one complete reasoning path from one entity to another entity, which may hinder the accuracy of relation extraction. To address this issue, this article proposes a novel method to capture this reasoning path from one entity to another entity, thereby better simulating reasoning skills to classify relation between two entities. Furthermore, we introduce an additional attention layer to summarize multiple reasoning paths for further enhancing the performance of the DocRE model. Experimental results on a large-scale document-level dataset show that the proposed approach achieved a significant performance improvement on a strong heterogeneous graph-based baseline.
ISSN:2375-4699
2375-4702
DOI:10.1145/3572898