The Hateful Memes Challenge: Detecting Hate Speech in Multimodal Memes

TL;DR

This paper introduces a challenging multimodal meme dataset for hate speech detection, emphasizing the difficulty of the task and the gap between current models and human performance.

Contribution

It presents a new benchmark dataset with challenging examples to evaluate multimodal hate speech detection models, highlighting the need for more sophisticated approaches.

Findings

State-of-the-art models perform significantly worse than humans

Unimodal models struggle with the dataset's challenging examples

The dataset reveals the difficulty of detecting hate speech in memes

Abstract

This work proposes a new challenge set for multimodal classification, focusing on detecting hate speech in multimodal memes. It is constructed such that unimodal models struggle and only multimodal models can succeed: difficult examples ("benign confounders") are added to the dataset to make it hard to rely on unimodal signals. The task requires subtle reasoning, yet is straightforward to evaluate as a binary classification problem. We provide baseline performance numbers for unimodal models, as well as for multimodal models with various degrees of sophistication. We find that state-of-the-art methods perform poorly compared to humans (64.73% vs. 84.7% accuracy), illustrating the difficulty of the task and highlighting the challenge that this important problem poses to the community.