Chatlytics
Data-trained game chat sentiment and trend dashboard.
[ Visit project โ ]Summary
An ML pipeline that reads raw in-game chat logs and turns them into sentiment scores and trending-topic alerts on a live dashboard.
Problem you're solving
Game studios and esports orgs generate huge volumes of chat data but have no fast way to know how players are actually feeling or what's blowing up in chat right now.
Target user
Game developers, community managers, and esports organizations monitoring player sentiment and chat trends at scale.
Competitors
General social-listening tools like Brandwatch and Sprout Social are built for Twitter and brand mentions, not tuned for the slang and shorthand specific to in-game chat.
Insight
Off-the-shelf sentiment models choke on gaming chat's slang and abbreviations โ a model fine-tuned specifically on game chat, paired with statistical spike detection, catches trends generic tools miss.
Solution
Cleans and normalizes raw chat text, fine-tunes DistilBERT for multi-class sentiment classification on a gaming chat dataset (Dota chat via Kaggle), flags trending terms with Z-score spike detection, and surfaces it all on an interactive dashboard.
- Python
- PyTorch
- Hugging Face Transformers
- pandas
- scikit-learn
- Streamlit
Distribution
Open-sourced on GitHub as a portfolio and research project rather than distributed directly to studios.
Adapting to user feedback
As a research-driven build rather than a live product, the main iteration loop was technical โ swapping in a fine-tuned DistilBERT model after generic sentiment classifiers underperformed on gaming-specific slang.