Performance Evaluation of Table Structure Recognition by Large Language Models on Agriculture and Forestry Standards Technical Documents

Abstract

This study evaluates the performance of large language models (LLMs) in recognizing table structures from PDFs. Using the TOITA benchmark, we assessed how accurately GPT-4o and GPT-4o-mini converted tables into CSV format. Results show that while both models demonstrated moderate performance, Yomitoku identified as optimal in prior research still achieved the highest accuracy. These findings highlight the current limitations of LLMs in table recognition and suggest directions for future improvement.