CLAIMSPECT数据集
收藏资源简介:
CLAIMSPECT数据集是一组关于现实世界中的科学和政治声明的数据集,旨在展示CLAIMSPECT框架在处理复杂声明时的鲁棒性和准确性。该数据集由真实世界中的科学和政治声明及其对应的语料库组成,旨在帮助理解声明的多面性和验证声明的有效性。CLAIMSPECT通过将声明分解为方面和子方面,并从语料库中提取相关片段,来丰富每个方面的内容。此外,该数据集还提供了对声明各个方面不同观点的分析,以及它们各自的出现频率。该数据集的创建是为了帮助解决验证声明真实性的问题,并为相关领域的研究提供数据支持。
The CLAIMSPECT dataset is a curated collection of real-world scientific and political claims, designed to demonstrate the robustness and accuracy of the CLAIMSPECT framework when handling complex claims. Comprising these real-world claims and their corresponding corpora, the dataset is intended to facilitate understanding of the multifaceted nature of claims and the validation of their validity. The CLAIMSPECT framework enriches the content of each aspect by decomposing claims into aspects and sub-aspects, and extracting relevant segments from their associated corpora. Additionally, this dataset provides analyses of diverse perspectives on each aspect of the claims, along with their respective occurrence frequencies. This dataset was developed to help address the challenge of verifying the veracity of claims, and to provide data support for research in relevant fields.
ClaimSpect数据集概述
数据集背景
- 数据集为ACL 2025主会议论文《Beyond True or False: Retrieval-Augmented Hierarchical Analysis of Nuanced Claims》的官方实现数据。
- 主要研究 nuanced claims(微妙主张)的层次化分析,突破传统"真/假"二元判断框架。
核心特点
- 采用检索增强生成框架(ClaimSpect)
- 支持对科学和政治主张进行多层次解构:
- 将主张分解为可验证的方面和子方面
- 捕捉语料库中不同观点倾向(支持/中立/反对)
- 量化各观点的普遍性
数据构成
- 包含两类真实世界主张:
- 科学主张(疫苗比较等)
- 政治主张
- 数据构建流程:
- 主张生成
- 文献检索
- 文献下载
- 文献分块处理
数据获取
评估体系
- 提供多维度评估方案:
- 基线评估
- LLM-as-Judge评估
- 检索语料相关性检查
- 人工评估
- 人机对齐评估
引用信息
bibtex @inproceedings{ anonymous2025beyond, title={Beyond True or False: Retrieval-Augmented Hierarchical Analysis of Nuanced Claims}, author={Anonymous}, booktitle={Submitted to ACL Rolling Review - February 2025}, year={2025}, url={https://openreview.net/forum?id=6Io5Pmuh19}, note={under review} }

- 1Beyond True or False: Retrieval-Augmented Hierarchical Analysis of Nuanced Claims伊利诺伊大学厄巴纳-香槟分校计算机科学系 · 2025年



