OnToxKG: An ontology-based knowledge graph of toxic symbols and their manifestations

Online toxicity, evolving through complex multimodal items like memes, poses a significant challenge. Integrating expert knowledge into moderation systems is increasingly crucial for identifying the nuances of toxic symbology, particularly in memes. Despite the wealth of available expertise, its structured integration into automated systems for online content interpretation remains underdeveloped. This paper introduces the OnTox ontology and OnToxKG Knowledge Graph to address gaps in addressing online toxic symbology. OnTox defines the multimodal semantics of 799 potentially toxic symbols. OnToxKG, a multimodal knowledge graph, integrates these symbols with commonsense sources like Wikidata and WordNet. We demonstrate the practical applications of these resources in automatically analyzing meme toxicity. Screenshot displaying a cautionary message in red text: "Caution: This work includes toxic content that may cause psychological distress and does not represent the views or opinions of the authors." The background is white.

• View at Publisher
• View at Scopus