MSA at BEA 2025 Shared Task: Disagreement-Aware Instruction Tuning for Multi-Dimensional Evaluation of LLMs as Math Tutors

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Hikal, Baraa, Basem, Mohamed, Oshallah, Islam, Hamdi, Ali
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915303268548608
author Hikal, Baraa
Basem, Mohamed
Oshallah, Islam
Hamdi, Ali
author_facet Hikal, Baraa
Basem, Mohamed
Oshallah, Islam
Hamdi, Ali
contents We present MSA-MathEval, our submission to the BEA 2025 Shared Task on evaluating AI tutor responses across four instructional dimensions: Mistake Identification, Mistake Location, Providing Guidance, and Actionability. Our approach uses a unified training pipeline to fine-tune a single instruction-tuned language model across all tracks, without any task-specific architectural changes. To improve prediction reliability, we introduce a disagreement-aware ensemble inference strategy that enhances coverage of minority labels. Our system achieves strong performance across all tracks, ranking 1st in Providing Guidance, 3rd in Actionability, and 4th in both Mistake Identification and Mistake Location. These results demonstrate the effectiveness of scalable instruction tuning and disagreement-driven modeling for robust, multi-dimensional evaluation of LLMs as educational tutors.
format Preprint
id arxiv_https___arxiv_org_abs_2505_18549
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle MSA at BEA 2025 Shared Task: Disagreement-Aware Instruction Tuning for Multi-Dimensional Evaluation of LLMs as Math Tutors
Hikal, Baraa
Basem, Mohamed
Oshallah, Islam
Hamdi, Ali
Computation and Language
We present MSA-MathEval, our submission to the BEA 2025 Shared Task on evaluating AI tutor responses across four instructional dimensions: Mistake Identification, Mistake Location, Providing Guidance, and Actionability. Our approach uses a unified training pipeline to fine-tune a single instruction-tuned language model across all tracks, without any task-specific architectural changes. To improve prediction reliability, we introduce a disagreement-aware ensemble inference strategy that enhances coverage of minority labels. Our system achieves strong performance across all tracks, ranking 1st in Providing Guidance, 3rd in Actionability, and 4th in both Mistake Identification and Mistake Location. These results demonstrate the effectiveness of scalable instruction tuning and disagreement-driven modeling for robust, multi-dimensional evaluation of LLMs as educational tutors.
title MSA at BEA 2025 Shared Task: Disagreement-Aware Instruction Tuning for Multi-Dimensional Evaluation of LLMs as Math Tutors
topic Computation and Language
url https://arxiv.org/abs/2505.18549