A Multi-Modal Dataset for Hate Speech Detection on Social Media: Case-study of Russia-Ukraine Conflict

Hate speech consists of types of content (e.g. text, audio, image) that express derogatory sentiments and hate against certain people or groups of individuals. The internet, particularly social media and microblogging sites, have become an increasingly popular platform for expressing ideas and opinions. Hate speech is prevalent in both offline and online media. A substantial proportion of this kind of content is presented in different modalities (e.g. text, image, video). Taking into account that hate speech spreads quickly during political events, we present a novel multimodal dataset composed of 5680 text-image pairs of tweets data related to the Russia-Ukraine war and annotated with a binary class:”hate” or”no-hate” The baseline results show that multimodal resources are relevant to leverage the hateful information from different types of data. The baselines and dataset provided in this paper may boost researchers in direction of multimodal hate speech, mainly during serious conflicts such as war contexts.