AIChartClaim
收藏资源简介:
AIChartClaim是一个多模态科学数据集,由宾夕法尼亚州立大学等机构创建,专注于人工智能领域的声明验证。该数据集包含科学发现作为声明、带有文本标题的图表图像作为证据以及解释文本,旨在通过结合文本和视觉证据进行联合推理。数据集通过构建两层多模态图(文本层和视觉层)实现跨模态检索,并采用对比学习目标优化检索过程。其创新性在于填补了科学领域多模态可解释声明验证数据集的空白,可应用于科学事实核查、学术图表理解等场景,提升语言模型对科学图表中量化关系的推理能力。
AIChartClaim is a multimodal scientific dataset developed by institutions including Pennsylvania State University, focusing on artificial intelligence-driven claim verification. This dataset comprises scientific claims as statements, chart images with textual titles as supporting evidence, and explanatory texts, aiming to conduct joint reasoning by combining textual and visual evidence. It implements cross-modal retrieval by constructing a two-layer multimodal graph (text layer and visual layer), and optimizes the retrieval process with contrastive learning objectives. Its innovation lies in filling the gap of multimodal interpretable claim verification datasets in the scientific domain. It can be applied in scenarios such as scientific fact-checking and academic chart understanding, enhancing the reasoning ability of large language models for quantitative relationships in scientific charts.
MEVER数据集概述
数据集来源
- 数据集来源于EACL-26论文《MEVER: Multi-Modal and Explainable Claim Verification with Graph-based Evidence Retrieval》。
数据集内容
- 该数据集用于多模态且可解释的声明验证任务。
- 数据集支持基于图的证据检索方法。
相关资源
- 数据集与源代码一同提供,地址为:https://github.com/cezhang01/mever




