The Pleasure(s) of Griefing: Tracing Multimodal Toxicity in League of Legends

The paper examines multimodal toxicity in League of Legends, arguing that hostile communication in competitive online play is not confined to text chat but emerges from the combined use of pings, emotes, UI interactions, champion positioning and chat messages. Its central claim is that toxic episodes typically appear as clusters across these channels and that understanding them requires temporally synchronized, context-aware analysis rather than surface-level classification. The authors situate this within broader questions about how competitive intensity, communication norms and uneven experiences of enjoyment intersect in contemporary online games.

The primary contribution is a prototype pipeline called MuMoDeTox, which combines YOLOv8-based visual detection of pings and champion positions from video frames, minimap localization, OCR extraction of chat and on-screen text, and integration of external match metadata such as champion role and performance statistics. The system produces a structured, queryable representation of how visual, textual and spatial signals co-occur at each moment of a match. The paper presents this as a methodological and technical contribution, and it describes the pipeline's configuration rather than reporting large-scale validation results.

The authors outline six research directions the pipeline enables: temporal patterns of hostile signaling around events such as deaths, spatial targeting and blame on the minimap, cross-modal reinforcement, shifts in toxicity across early, mid and late game phases, qualitative close-up analysis informed by conversation analysis, and player-type-specific multimodal patterns. Initial observations from real gameplay suggest that the same communicative resources, such as repeated question-mark pings or line-broken chat messages, can serve antagonistic, strategic or playful functions depending on timing and context. The paper frames this as evidence that toxicity reflects both situational pressure and stable communicative habits, though these observations are preliminary and not drawn from a formally reported dataset.