Many business workflows require extracting important fields from form-like documents (e.g. bank statements, bills of lading, purchase orders, etc.). Recent techniques for automating this task work well only when trained with large datasets. In this work we propose a novel data augmentation technique to improve performance when training data is scarce, e.g. 10-250 documents. Our technique, which we call FieldSwap, works by swapping out the key phrases of a source field with the key phrases of a target field to generate new synthetic examples of the target field for use in training. We demonstrate that this approach can yield 1-7 F1 point improvements in extraction performance.
@article{arxiv.2212.10047,
title = {An Augmentation Strategy for Visually Rich Documents},
author = {Jing Xie and James B. Wendt and Yichao Zhou and Seth Ebner and Sandeep Tata},
journal= {arXiv preprint arXiv:2212.10047},
year = {2022}
}