HarMeme

该数据集包含了在美国社交媒体上发布的与新冠疫情相关的梗图，这些梗图均附有通过谷歌视觉API生成的相关文本内容，并按照是否含有仇恨性质被分为两个类别：仇恨和非仇恨。此外，该数据集被划分为85%的训练集、5%的验证集以及10%的测试集。数据集规模为3,544张梗图，其任务是进行仇恨梗图检测。

This dataset contains COVID-19-related memes posted on U.S. social media platforms. Each meme is paired with relevant textual content generated via the Google Vision API, and is classified into two categories based on whether it contains hateful content: hate and non-hate. Additionally, the dataset is split into an 85% training set, 5% validation set, and 10% test set. Comprising a total of 3,544 memes, this dataset is designed for the task of hate meme detection.

搜集汇总

背景与挑战

背景概述

HarMeme是一个多模态模因数据集，包含图像和标注信息，需在特定技术环境下使用MMF框架运行。