This is a project that introduces a harmful meme dataset, which for the first time adds CoT annotation to enable the model to mimic human thinking to infer whether memes are harmful.
This work is introduced in our paper: MemeMind: A Large-Scale Multimodal Dataset with Chain-of-Thought Reasoning for Harmful Meme Detection.
You can view and download our dataset MemeMind through this link: https://huggingface.co/datasets/yubb66/MemeMind. This includes all harmful meme samples and annotation files in the dataset, divided into training and testing sets.
The supplementary materials for our paper have been provided in detail in the Supplementary folder, including reference, appendices, annotated visualization examples of the dataset, and visualization examples of the output results of our method model.
The relevant code for our method will be open sourced soon after we have organized it.