LLM analyzes Twitch chats, finds 2.4% toxicity rate

Researchers have analyzed approximately 20 million chat messages from Twitch streams across seven game genres to understand toxicity patterns. Using a pre-trained Large Language Model for zero-shot classification, they found that 2.4% of messages were toxic, with MOBA games showing the highest rate (3.2%) and sports games the lowest (2%). The study highlights significant differences in toxicity even within genres, suggesting game-specific community norms influence behavior and informing targeted moderation strategies. AI

IMPACT Provides insights into LLM capabilities for content moderation and understanding online community behavior.

RANK_REASON Academic paper analyzing toxicity using LLMs. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• Large Language Model
• Twitch

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →