RE$^2$: Region-Aware Relation Extraction from Visually Rich Documents

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Ramu, Pritika, Wang, Sijia, Mouatadid, Lalla, Rimchala, Joy, Huang, Lifu
Natura: Preprint
Pubblicazione: 2023
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866916271978708992
author Ramu, Pritika
Wang, Sijia
Mouatadid, Lalla
Rimchala, Joy
Huang, Lifu
author_facet Ramu, Pritika
Wang, Sijia
Mouatadid, Lalla
Rimchala, Joy
Huang, Lifu
contents Current research in form understanding predominantly relies on large pre-trained language models, necessitating extensive data for pre-training. However, the importance of layout structure (i.e., the spatial relationship between the entity blocks in the visually rich document) to relation extraction has been overlooked. In this paper, we propose REgion-Aware Relation Extraction (RE$^2$) that leverages region-level spatial structure among the entity blocks to improve their relation prediction. We design an edge-aware graph attention network to learn the interaction between entities while considering their spatial relationship defined by their region-level representations. We also introduce a constraint objective to regularize the model towards consistency with the inherent constraints of the relation extraction task. Extensive experiments across various datasets, languages and domains demonstrate the superiority of our proposed approach.
format Preprint
id arxiv_https___arxiv_org_abs_2305_14590
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle RE$^2$: Region-Aware Relation Extraction from Visually Rich Documents
Ramu, Pritika
Wang, Sijia
Mouatadid, Lalla
Rimchala, Joy
Huang, Lifu
Computation and Language
Artificial Intelligence
Current research in form understanding predominantly relies on large pre-trained language models, necessitating extensive data for pre-training. However, the importance of layout structure (i.e., the spatial relationship between the entity blocks in the visually rich document) to relation extraction has been overlooked. In this paper, we propose REgion-Aware Relation Extraction (RE$^2$) that leverages region-level spatial structure among the entity blocks to improve their relation prediction. We design an edge-aware graph attention network to learn the interaction between entities while considering their spatial relationship defined by their region-level representations. We also introduce a constraint objective to regularize the model towards consistency with the inherent constraints of the relation extraction task. Extensive experiments across various datasets, languages and domains demonstrate the superiority of our proposed approach.
title RE$^2$: Region-Aware Relation Extraction from Visually Rich Documents
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2305.14590