Field Extraction from Forms with Unlabeled Data

TL;DR

This paper introduces a new framework for extracting form fields from unlabeled data by leveraging pseudo-labels and a transformer model, improving accuracy through a progressive refinement process.

Contribution

It presents a novel combination of rule-based pseudo-label mining and a transformer-based extraction model with a refinement module for unlabeled form data.

Findings

Effective field extraction demonstrated on unlabeled forms

Pseudo-label refinement improves model robustness

Transformer-based representations enhance extraction accuracy

Abstract

We propose a novel framework to conduct field extraction from forms with unlabeled data. To bootstrap the training process, we develop a rule-based method for mining noisy pseudo-labels from unlabeled forms. Using the supervisory signal from the pseudo-labels, we extract a discriminative token representation from a transformer-based model by modeling the interaction between text in the form. To prevent the model from overfitting to label noise, we introduce a refinement module based on a progressive pseudo-label ensemble. Experimental results demonstrate the effectiveness of our framework.