Toxicity_Detection_Presentation_SID.pptx

Toxicity_Detection_Presentation_SID.pptx

• 1.
  z Can AI Keep OnlineGaming Civil? How Machines Learn to Detect Toxicity — and What It Says About Us By Siddharth Rao Kartik | CMPE 255 – Data Mining Short Story Assignment
• 2.
  z Introduction  Online gamestoday are massive social platforms where millions of players communicate every minute. Toxic chat behavior, however, is a growing concern. Human moderation alone cannot handle the scale, which is why researchers are turning to AI to detect harmful messages automatically.
• 3.
  z Research Focus  The study'A Comparative Study on Toxicity Detection in Gaming Chats Using Machine Learning and Large Language Models' by Yehor Tereshchenko and Mika Hämäläinen (2025) compares multiple AI techniques for moderation:  • Traditional ML models using TF-IDF and embeddings  • Fine-tuned transformer models such as DistilBERT  • Large Language Models (LLMs) for zero/few-shot learning
• 4.
  z Key Findings  DistilBERT achievedthe best trade off ‑ between accuracy, processing speed, and computational cost. Traditional models were fast but less context- aware, while LLMs performed well but were too expensive and slow for real- time moderation.
• 5.
  z Relevance to Data Mining This study demonstrates several core data mining principles:  • Text classification and pattern recognition  • Dimensionality reduction via embeddings  • Clustering of behavior and anomaly detection  • Evaluation metrics such as precision, recall, and F1-score
• 6.
  z Ethical and Practical Insights Automated moderation must balance fairness and empathy. Mislabeling harmless banter can alienate players, while missing genuine abuse harms communities. A hybrid approach—AI for filtering, humans for review—offers the best path forward.
• 7.
  z Conclusion  This researchhighlights how data mining and NLP can support healthier online communities. Combining efficiency of AI with human understanding ensures fairness and scalability in gaming moderation.
• 8.
  z z Thank you