RE$^2$: Region-Aware Relation Extraction from Visually Rich Documents
This addresses form understanding for document processing applications, representing an incremental improvement over existing methods.
The paper tackles the problem of relation extraction from visually rich documents by leveraging region-level spatial structure between entity blocks, achieving state-of-the-art performance across multiple datasets, languages, and domains.
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.