Data-Gouv-FR/etat-de-la-mesinformation-et-desinformation-climatique-dans-les-medias-audiovisuels-en-2025
收藏资源简介:
该数据集包含与气候错误信息和虚假信息相关的文件,这些文件由法国媒体生态观察站在开放数据平台上发布。数据集是法国媒体生态观察站和Science Feedback合作的成果,后者是法国气候主题事实核查的权威组织,合作项目为Climate Safegards。整个过程还得到了专家委员会的支持。数据允许对2025年1月以来视听媒体中气候错误信息的演变进行盘点,研究事实核查员识别的案例在不同媒体之间的分布,并观察主要的气候虚假信息叙事。数据集包含以下文件:evolution_of_misinformation.csv、misinformation_per_media和disinformation_narratives_with_debunks。方法论前提中,气候虚假信息被定义为虚假或误导性言论,高风险误导公众关于气候变化和气候行动的事实,这些事实基于IPCC确立的科学知识;气候错误信息则区别在于说话者没有明显的伤害意图,可能源于错误或对误导性叙事的渗透。数据处理步骤包括:通过API获取原始数据(电视/广播流的视听转录)、识别处理气候变化的序列(基于主题关键词词典的方法)、检测可能包含气候错误信息的案例(基于事实核查员手动标记数据微调的AI模型)、对所有检测到的案例进行手动事实核查以确保数据稳健性、将案例与重复的虚假信息叙事关联(如果>8个案例对应同一叙事)。数据集将每3到6个月更新一次,具体取决于法国媒体生态观察站的资源可用性。
The dataset contains files related to climate misinformation and disinformation, published as open data on the French Media Ecology Observatory website. It is the result of a collaboration between the French Media Ecology Observatory and Science Feedback, a leading organization in France for fact-checking on climate topics, as part of the Climate Safeguards project. The entire process was also supported by an expert committee. The data allows for an assessment of the evolution of climate misinformation in audiovisual media since January 2025, studying the distribution of cases identified by fact-checkers across different media, and observing the prevalent narratives of climate disinformation. The dataset includes the following files: evolution_of_misinformation.csv, misinformation_per_media, and disinformation_narratives_with_debunks. Methodologically, climate disinformation is defined as false or misleading discourse with a high risk of misleading the public on facts established by scientific knowledge about climate change and climate action regarding mitigation and adaptation measures as established by the IPCC. Climate misinformation is distinguished by the absence of a demonstrated intent to harm, possibly due to error or susceptibility to misleading narratives. Data processing steps include: acquiring raw data (audiovisual transcriptions of TV/radio streams) via API, identifying sequences addressing climate change (based on thematic keyword dictionaries), detecting cases at risk of containing climate misinformation (using an AI model fine-tuned on manually labeled data by fact-checkers), manual fact-checking of all detected cases to ensure data robustness, and associating cases with recurring disinformation narratives (if >8 cases correspond to the same narrative). The dataset will be updated every 3 to 6 months, depending on resource availability for the French Media Ecology Observatory.




