IEEE CoG 2019 - Industry Day - Marko Knezevic (Nordeus)
summary
Weekly recommended-player feature replaces rule heuristics with model-based collaborative filtering trained on about 200 features, using token-thresholded purchases as true signals and under-sampling to fix class imbalance, plus a rule-based fallback for cold-starts. Engineering limits predictions to highly active users to cut latency. A/B testing shows a clear lift — about 30% uplift in recommended-player purchases — and a matching rise in ARPDAU, confirming the model increases valuable buys rather than just shifting existing spend.
AI-written summary · about
chapters (7)
- 0:00Introduction to Nordeus
- 2:55Personalizing player recommendations
- 8:48Model selection and training
- 13:29Handling data challenges
- 15:18A/B test results and success
- 17:06Key takeaways and conclusion
- 18:29Q&A on player reactions
related talks
23mCustom-made games with machine learning and big data
28mThe Power of Personalized Customer Experiences Through AI Innovations
27mEnhancing Personalization Using Recommendation Systems & Multi-Armed Bandits
20mUsers acquired - Now what?: Maximising revenue from existing players
36mHow to Enhance the Gaming Experience Through Content Micro-personalisation
43mRecommendation Systems in Market Research
20mDataEngBytes 2023 - PER-T1-06 - Kamal Hossain
30mOptimizing Mobile Game LiveOps with Data-Driven Decisions