GeoEval : benchmark for evaluating LLMs and multi-modal models on geometry problem-solving

Zhang, Jiaxin and Li, Zhong Zhi and Zhang, Ming Liang and Yin, Fei and Liu, Cheng Lin and Moshfeghi, Yashar; Ku, Lun-Wei and Martins, Andre and Srikumar, Vivek, eds. (2024) GeoEval : benchmark for evaluating LLMs and multi-modal models on geometry problem-solving. In: 62nd Annual Meeting of the Association for Computational Linguistics, ACL 2024. Proceedings of the Annual Meeting of the Association for Computational Linguistics . Association for Computational Linguistics (ACL), THA, pp. 1258-1276. ISBN 9798891760998 (https://doi.org/10.18653/v1/2024.findings-acl.73)

[thumbnail of Zhang-etal-ACL-Benchmark-for-Evaluating-LLMs-and-Multi-Modal-Models-on-Geometry-Problem-Solving]
Preview
 Text. Filename: Zhang-etal-ACL-Benchmark-for-Evaluating-LLMs-and-Multi-Modal-Models-on-Geometry-Problem-Solving.pdf
Final Published Version
License: Creative Commons Attribution 4.0 logo
Download (5MB)| Preview

Abstract

Recent advancements in large language models (LLMs) and multi-modal models (MMs) have demonstrated their remarkable capabilities in problem-solving. Yet, their proficiency in tackling geometry math problems, which necessitates an integrated understanding of both textual and visual information, has not been thoroughly evaluated. To address this gap, we introduce the GeoEval benchmark, a comprehensive collection that includes a main subset of 2,000 problems, a 750 problems subset focusing on backward reasoning, an augmented subset of 2,000 problems, and a hard subset of 300 problems. This benchmark facilitates a deeper investigation into the performance of LLMs and MMs in solving geometry math problems. Our evaluation of ten LLMs and MMs across these varied subsets reveals that the WizardMath model excels, achieving a 55.67% accuracy rate on the main subset but only a 6.00% accuracy on the hard subset. This highlights the critical need for testing models against datasets on which they have not been pre-trained. Additionally, our findings indicate that GPT-series models perform more effectively on problems they have rephrased, suggesting a promising method for enhancing model capabilities.

ORCID iDs

Zhang, Jiaxin ORCID logoORCID: https://orcid.org/0000-0001-7355-7975, Li, Zhong Zhi, Zhang, Ming Liang, Yin, Fei, Liu, Cheng Lin and Moshfeghi, Yashar ORCID logoORCID: https://orcid.org/0000-0003-4186-1088; Ku, Lun-Wei, Martins, Andre and Srikumar, Vivek
CORE (COnnecting REpositories)