Lune

ACL2025顶会

MCS-Bench: A Comprehensive Benchmark for Evaluating Multimodal Large Language Models in Chinese Classical Studies

Yang Liu, Jiahuan Cao, Hiuyi Cheng, Yongxin Shi, Kai Ding, Lianwen Jin

2025年份
2顶会引用

摘要

With the rapid development of Multimodal Large Language Models (MLLMs), their potential in Chinese Classical Studies (CCS), a field which plays a vital role in preserving and promoting China's rich cultural heritage, remains largely unexplored due to the absence of specialized benchmarks. To bridge this gap, we propose MCS-Bench, the first-of-its-kind multimodal benchmark specifically designed for CCS across multiple subdomains. MCS-Bench spans seven core subdomains (Ancient Chinese Text, Calligraphy, Painting, Oracle Bone Script, Seal, Cultural Relic, and Illustration), with a total of 45 meticulously designed tasks. Through extensive evaluation of 37 representative MLLMs, we observe that even the topperforming model (InternVL2.5-78B) achieves an average score below 50, indicating substantial room for improvement. Our analysis reveals significant performance variations across different tasks and identifies critical challenges in areas such as Optical Character Recognition (OCR) and cultural context interpretation. MCS-Bench not only establishes a standardized baseline for CCS-focused MLLM research but also provides valuable insights for advancing cultural heritage preservation and innovation in the Artificial General Intelligence (AGI) era. Data and code will be publicly available. Question Fromat Method Dataset Domain Modality License Scale # Category # Task # LLM MCQ QA HG CI MC C-Eval General Text-only CC BY-NC-SA-4.0 439 1 2 11 Chinese SimpleQA General Text-only CC BY-NC-SA-4.0 323 4 11 41 CIF-Bench General Text-only -150 1 3 28 CMMLU General Text-only CC BY-NC-4.0 1,192 1 7 21 GAOKAO-Bench General Text-only Apache-2.0 145 1 2 12 XiezhiBenchmark General Text-only CC BY-NC-SA-4.0 2,060 2 3 47 ACLUE CCS Text-only CC BY-NC-4.0 4,967 5 15 8 C-CLUE CCS Text-only CC BY-SA-4.0 1,122 1 2 -CCLUE CCS Text-only Apache-2.0 36,319 2 5 -CCPM CCS Text-only -2,720 1 1 -THUAIPoet CCS Text-only -5,173 1 3 -WenMind CCS Text-only CC BY-NC-SA-4.0 4,875 3 42 31 WYWEB CCS Text-only -69,700 5 9 -CII-Bench General Image-Text Apache-2.0 137 1 1 21 ALM-Bench Culture Image-Text CC BY-NC-4.0 466 18 -16 CulturalBench Culture Image-Text CC BY-4.0 117 3 -18 CVQA Culture Image-Text -311 10 -16 MaRVL Culture Image-Text CC BY-4.0 1,012 11 1 -MCS-Bench (Ours) CCS Image-Text CC BY-NC-SA-4.0 6,500 7 45 37

问问这篇 Paper

智能体会读完全文。

Lune 把这篇 Paper 索引到了每一个公式,引用它的顶会 Paper 也一样。你提问,回答直接引用原文。

可以从这些问题问起

智能体调用

Luneget_paper_fulltext

在 Lune 里问

免费开始,无需绑卡

引用它的顶会 Paper2

问问它们各自怎么用它

它引用的顶会 Paper6

相关 Paper

黄昏的海面,两侧是细线勾勒的悬崖