Bengali Hateful Memes Detection: A Comprehensive Dataset and Deep Learning Approach

The rise of hateful memes on social media, which promote racism, misogyny, and discrimination, demands advanced detection techniques. These memes can normalize prejudice and negatively impact public opinion. This study focuses on creating enhanced auto-matic detection methods for such content, utilizing five pre-trained deep learning models: VGG16, ResNet50, InceptionV3, DenseNet121, and Xception. We tested these models on a dataset of 3000 Bengali memes. The Xception model showed the best performance, achieving an accuracy of 68.35%. This research contributes to efforts in combating harmful content's spread on social media..