Improving Model Evaluation using SMART Filtering of Benchmark Datasets

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Gupta, Vipul, Ross, Candace, Pantoja, David, Passonneau, Rebecca J., Ung, Megan, Williams, Adina
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910821361123328
author Gupta, Vipul
Ross, Candace
Pantoja, David
Passonneau, Rebecca J.
Ung, Megan
Williams, Adina
author_facet Gupta, Vipul
Ross, Candace
Pantoja, David
Passonneau, Rebecca J.
Ung, Megan
Williams, Adina
contents One of the most challenging problems facing NLP today is evaluation. Some of the most pressing issues pertain to benchmark saturation, data contamination, and diversity in the quality of test examples. To address these concerns, we propose Selection Methodology for Accurate, Reduced, and Targeted (SMART) filtering, a novel approach to select a high-quality subset of examples from existing benchmark datasets by systematically removing less informative and less challenging examples. Our approach applies three filtering criteria, removing (i) easy examples, (ii) data-contaminated examples, and (iii) examples that are similar to each other based on distance in an embedding space. We demonstrate the effectiveness of SMART on three multiple choice QA datasets, where our methodology increases efficiency by reducing dataset size by 48\% on average, while increasing Pearson correlation with rankings from ChatBot Arena, a more open-ended human evaluation setting. Our method enables us to be more efficient, whether using SMART to make new benchmarks more challenging or to revitalize older datasets, while still preserving the relative model rankings.
format Preprint
id arxiv_https___arxiv_org_abs_2410_20245
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Improving Model Evaluation using SMART Filtering of Benchmark Datasets
Gupta, Vipul
Ross, Candace
Pantoja, David
Passonneau, Rebecca J.
Ung, Megan
Williams, Adina
Computation and Language
Artificial Intelligence
Machine Learning
One of the most challenging problems facing NLP today is evaluation. Some of the most pressing issues pertain to benchmark saturation, data contamination, and diversity in the quality of test examples. To address these concerns, we propose Selection Methodology for Accurate, Reduced, and Targeted (SMART) filtering, a novel approach to select a high-quality subset of examples from existing benchmark datasets by systematically removing less informative and less challenging examples. Our approach applies three filtering criteria, removing (i) easy examples, (ii) data-contaminated examples, and (iii) examples that are similar to each other based on distance in an embedding space. We demonstrate the effectiveness of SMART on three multiple choice QA datasets, where our methodology increases efficiency by reducing dataset size by 48\% on average, while increasing Pearson correlation with rankings from ChatBot Arena, a more open-ended human evaluation setting. Our method enables us to be more efficient, whether using SMART to make new benchmarks more challenging or to revitalize older datasets, while still preserving the relative model rankings.
title Improving Model Evaluation using SMART Filtering of Benchmark Datasets
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2410.20245