Detecting Hate and Inflammatory Content in Bengali Memes: A New Multimodal Dataset and Co-Attention Framework

The article introduces the Bn-HIB (Bangla Hate Inflammatory Benign) dataset, comprising 3,247 annotated Bengali memes categorized as Benign, Hate, or Inflammatory, marking the first dataset to differentiate inflammatory content from direct hate speech in this language. It also presents the Multi-Modal Co-Attention Fusion Model (MCFM), which utilizes a co-attention mechanism to analyze and integrate visual and textual features for improved classification accuracy, outperforming existing models on the Bn-HIB dataset. This work is significant for practitioners as it addresses the detection of harmful content in low-resource languages, providing both a valuable dataset and an effective model architecture for further research and application.