Reading Order Matters: Information Extraction from Visually-rich Documents by Token Path Prediction

TL;DR

This paper introduces Token Path Prediction (TPP), a novel approach for extracting entities from visually-rich documents that overcomes reading order issues affecting traditional sequence-labeling methods.

Contribution

The paper proposes TPP, modeling document layout as a token graph and predicting entity paths, along with revised benchmarks for realistic evaluation of VrD-NER systems.

Findings

TPP outperforms traditional BIO-tagging methods in VrD-NER tasks.

Revised benchmarks better reflect real-world OCR scenarios.

TPP shows potential as a universal solution for document information extraction.

Abstract

Recent advances in multimodal pre-trained models have significantly improved information extraction from visually-rich documents (VrDs), in which named entity recognition (NER) is treated as a sequence-labeling task of predicting the BIO entity tags for tokens, following the typical setting of NLP. However, BIO-tagging scheme relies on the correct order of model inputs, which is not guaranteed in real-world NER on scanned VrDs where text are recognized and arranged by OCR systems. Such reading order issue hinders the accurate marking of entities by BIO-tagging scheme, making it impossible for sequence-labeling methods to predict correct named entities. To address the reading order issue, we introduce Token Path Prediction (TPP), a simple prediction head to predict entity mentions as token sequences within documents. Alternative to token classification, TPP models the document layout as a complete directed graph of tokens, and predicts token paths within the graph as entities. For better evaluation of VrD-NER systems, we also propose two revised benchmark datasets of NER on scanned documents which can reflect real-world scenarios. Experiment results demonstrate the effectiveness of our method, and suggest its potential to be a universal solution to various information extraction tasks on documents.