Digital gaming is now an everyday activity and is recognized as a cultural asset, with games displayed at MoMA and Esports featured at the 2021 Olympics. Among these, League of Legends (LoL) has over 150 million players, with more than 2 million active at any time. LoL is known for its toxic community, where players often experience harassment and other negative behaviors. Efforts to mitigate toxicity, like automatic harmful chat detection, had limited success, as players offend through both visual and textual channels.

Our project aims to detect toxic communication in LoL by analyzing multimodal interactions (verbal and non-verbal) in relation to in-game events. Using machine learning, we will extract and assess chat and visual symbols ("pings"), which are misused to express hostility. Pings have community-specific meanings and can become toxic in particular contexts; by tracking their location and timing relative to player actions, we aim to detect toxic behavior early in gameplay.