Who’s Hated: Detecting and Analyzing the Entities Targeted by Hateful Memes

Hateful memes pose a significant threat to the well-being of online communities. Therefore, developing automated systems for the detection and analysis of hateful memes is crucial to mitigate their adverse impact; nonetheless, it is an intrinsically difficult and open problem: memes convey messages using both images and texts and, hence, require multimodal reasoning. While previous research has examined similar problems, they are quite limited; a holistic approach is lacking, particularly in terms of reasoning about the target entities. Moreover, there is little analysis that clarifies why certain entities are more susceptible, and no suggested measures have been put forth to specifically curb the dissemination of hateful memes. In this study, we aim to address these issues.