Unveiling Hate: Multimodal Perspectives and Knowledge Graphs

Deep learning models continue to face challenges due to the growing amount of hate speech spread online. The recent advances in research have led to the development of several models that may be utilised to address these difficulties. This covers both unimodal and multimodal models; however, due to accuracy limits and performance restrictions, a more sophisticated method is required for these kinds of issues. This study evaluates the effectiveness of previously proposed models and frameworks for the identification of hateful memes and makes recommendations for how to improve them to increase accuracy. The significance of multimodal techniques for categorising multimodal content—like memes, which include text and picture features—is also highlighted in the research. Knowledge graphs are used in the proposed approach to solve the problem of damaging analogies and stereotypes found in memes. External world information is also used to create multimodal models that effectively categorise memes that are closely connected to real-world events and data.