Multimodal Hate Speech Detection With Explanability Using LIME

Social media platforms link users from all over the world and enable opinion exchange on subjects including politics, finances and social issues. Policies governing objectionable material such as hate speech, that could potentially cause users psychological harm are difficult to reliably implement. Social media platform creators typically rely on human moderators to look into and assess potentially harmful content. Artificial intelligence (AI)-based decision support methods have attracted a lot of attention recently in regard to the identification of hate speech. For instance, AI-based models can be used to identify a variety of undesirable speech communication concepts, such as racism, hate speech and inflammatory language. This paper aims to develop an Explainable AI model for the automatic detection of hate speech in online content, providing transparency and interpretability in the decision-making process. Our initiative addresses the critical issue of hate speech in social media platforms through a multi-modal approach, encompassing both text and image analysis. Building a robust machine learning model that can precisely identify hate speech in diverse content types is one of our key objectives. To foster openness and trust, we employ sophisticated explainability approaches that provide a comprehensive explanation of the reasoning underlying the model’s predictions for both text and image inputs. This ensures that users have a clear understanding of how the model makes decisions, enhancing transparency in our detection system. By making the internet a safer and friendlier environment for everyone, our multi-modal approach enhances the overall effectiveness of combating hate speech across different content formats.