Alisa Chumachenko, GOSU

Gaming Data Magic

(White Nights Conference Berlin 2019)
The official conference website — http://wnconf.com

Transcript

1. FUTURE OF PERSONALIZATION IN GAMES Alisa Chumachenko for White Nights 2019
2. Alisa • 17 years in gaming • Game Insight founder and ex-CEO (2009-2015) • GOSU Data Lab Founder 
 (2017-2019) • In Top 30 women changed IT by Techcrunch • 50+ titles launched
4. PUBG
5. CS:GO
6. DOTA 2
7. LEAGUE OF LEGEND CHAMPIONS
9. Mold of unique gamers behaviors • Identification • Esports (Online Tournaments) • Matchmaking • Personal features • SmartBots GAMER FINGERPRINT
10. Machine learning approach More skilled bots: model plays millions of games, each game outcome is taken into account in the learning process to choose the most effective actions
 More human bots: model predicts which action person would have made SMART BOTS/NPC C A S E S T U DY ALPHA GO STARCRAFT
12. Traditional Analytics approach • Balanced game — good game • Rating (MMR) MATCHMAKING Machine learning approach • The selection of the relevant group based on the user experience and game preferences; • Multi-factor analysis of player`s skills and involvement in the gameplay Beyond Skill Rating: Advanced Matchmaking in Ghost Recon Online” Olivier Delalleau et. al. C A S E S T U DY
13. GOSU PARTY FINDER C A S E S T U DY On the basis of millions analyzed matches, AI suggests the most relevant potential party members for each GOSU.AI user
14. Traditional analytics approach • Send special activation offer of award on a Nth day after player stops playing SMART CHURN PREVENTION Machine learning approach • Train machine learning model that predicts probability of user not to do any action in next week based on previous data.
 • Send activation offer when predicted probability falls below the threshold, not after fixed period of time C A S E S T U DY
15. Traditional analytics approach • Users segmentation by clusters; • Selection of triggers /conditions for each cluster. PERSONAL OFFER/
 PRICING Machine learning approach • Data collection: random sentences on a small target group; • Model training; • Decision making: model using for each player individually • Send offer at the time when the user is ready enough for accepting the offer (smart trigger / when probability of acceptance is high enough). C A S E S T U DY
16. Machine learning approach: • Predict user preferences using pre-trained machine learning model (...preference examples)
 • Change gameplay according to predicted preferences, i.e. more pvp quests for “brutal" man, more crafting quests and beautiful skins for a tender girl DYNAMIC GAMEPLAY (PERSONALIZATION AND PERSONAL RECOMMENDATIONS) C A S E S T U DY
18. ALISA CHUMACHENKO a@gosu.ai THANKS!