QuotaClimat/frugalaichallenge-text-train

该数据集是为2025年Frugal AI Challenge设计的，旨在通过跟踪AI模型的能源消耗和性能，鼓励高效部署AI模型。数据集包含约6000条与气候相关的引用和声明，专注于识别和分类气候虚假信息。数据集结合了来自电视、广播和在线平台的各种媒体来源的引用和声明，帮助训练模型识别不同类型的气候虚假信息。数据集的结构包括文本和标签，标签基于简化的CARDS分类法，共有7个主要类别。数据集分为训练集和测试集，测试集为挑战赛隐藏。数据集的创建过程包括从DeSmog气候虚假信息数据库和FLICC数据集中提取和注释数据，并经过GPT4o-mini和手动验证。数据集的使用许可为CC BY-NC 4.0。

The dataset has been built for the Frugal AI Challenge 2025, aimed at encouraging both academic and industry actors to keep efficiency in mind when deploying AI models. It contains approximately ~6000 climate-related quotes and statements, specifically focused on identifying and categorizing climate disinformation narratives. The dataset combines quotes and statements from various media sources to help train models that can identify different types of climate disinformation claims. The labels are based on a simplified version of the CARDS taxonomy, containing 7 main labels. The dataset is split into training and testing sets, with a hidden test set for the challenge. The creation of the dataset combines data from two main sources curated by the QuotaClimat & Data For Good team, and has been validated with GPT4o-mini and manual annotations.