Extracting Information from Scientific Literature via Visual Table Question Answering Models

TL;DR

This paper evaluates methods for extracting information from scientific tables to improve question answering, emphasizing the importance of preserving table structure for accuracy in automated literature review tools.

Contribution

It compares OCR, pre-trained visual question answering models, and table structure recognition, highlighting the superiority of structure-preserving approaches for scientific data extraction.

Findings

Structure-preserving methods outperform others.

Accurate recognition of notations improves results.

Table structure integrity is crucial for reliable extraction.

Abstract

This study explores three approaches to processing table data in scientific papers to enhance extractive question answering and develop a software tool for the systematic review process. The methods evaluated include: (1) Optical Character Recognition (OCR) for extracting information from documents, (2) Pre-trained models for document visual question answering, and (3) Table detection and structure recognition to extract and merge key information from tables with textual content to answer extractive questions. In exploratory experiments, we augmented ten sample test documents containing tables and relevant content against RF- EMF-related scientific papers with seven predefined extractive question-answer pairs. The results indicate that approaches preserving table structure outperform the others, particularly in representing and organizing table content. Accurately recognizing specific notations and symbols within the documents emerged as a critical factor for improved results. Our study concludes that preserving the structural integrity of tables is essential for enhancing the accuracy and reliability of extractive question answering in scientific documents.