Friday 31 January 2025
The internet has become a breeding ground for hate speech, and memes have emerged as a popular way to spread harmful stereotypes and prejudices. However, identifying and countering these forms of online hate is a challenging task, especially when it comes to anti-Muslim sentiment.
Researchers from United International University in Bangladesh have developed a novel dataset and classifier specifically designed to detect and classify anti-Muslim hate speech in memes. The team’s approach uses the Vision-and-Language Transformer (ViLT) model, which integrates both visual and textual representations of memes to identify subtle forms of Islamophobic content.
The researchers collected a dataset of 953 memes from various online platforms, including Reddit, X, 9GAG, and Google Images. Each meme was carefully labeled as hateful or non-hateful towards Muslims by a team of annotators with experience in hate speech detection. The dataset included a range of examples, from overtly discriminatory content to more subtle forms of prejudice.
The researchers then trained the ViLT model on this dataset, using a combination of image and text features to identify patterns that distinguish hateful from non-hateful memes. The model achieved impressive results, with an F1-weighted score of 0.738 via 10-fold cross-validation. This suggests that the model is highly effective at identifying anti-Muslim hate speech in memes.
The study’s findings have important implications for online content moderation. By developing a classifier specifically designed to detect anti-Muslim hate speech in memes, researchers can help identify and remove harmful content from social media platforms. This could potentially reduce the spread of Islamophobic stereotypes and create a more inclusive online environment.
The study’s authors also highlight the challenges posed by subtle forms of hate speech, which often rely on coded language or visual cues to convey discriminatory messages. The ViLT model’s ability to integrate both visual and textual features allows it to identify these subtle forms of prejudice, making it a valuable tool for content moderation.
Overall, this research demonstrates the potential for AI-powered approaches to detect and counter online hate speech. By developing more sophisticated classifiers like the ViLT model, researchers can help create a safer and more inclusive online environment for all users.
Cite this article: “Detecting Anti-Muslim Hate Speech in Memes Using AI”, The Science Archive, 2025.
Hate Speech, Anti-Muslim Sentiment, Memes, Online Hate, Vilt Model, Vision-And-Language Transformer, Dataset, Classifier, Content Moderation, Islamophobic Stereotypes.
