- This event has passed.
Playing with data: How to use artificial intelligence to optimize video games
March 5, 2018 @ 3:00 pm - 4:00 pm
Public talk with Dr. África Periáñez
The video-game industry is experiencing a major revolution. Nowadays, games are always connected to the Internet and developers have at their disposal a huge amount of data, which can be used to try to increase player engagement and game monetization through big data analytics. In order to do so, they can employ suitable machine learning methods to model and predict player behavior.
In particular, by using sophisticated churn prediction models, developers can now predict when and where (i.e. in which part of the game) individual players are going to stop playing and take preventive actions to try to retain them. Survival analysis focuses on predicting when a certain event will happen, considering censored data. In this case, our event of interest is churn, and highly accurate prediction results can be obtained by combining survival models and ensemble learning techniques. On the other hand, and even though they do not take into account censoring of churn data, deep learning methods can also be helpful to foresee when a player will quit the game. In particular, long short-term memory networks –recurrent neural networks trained using backpropagation through time– constitute an excellent approach to model sequential problems.
In this talk, I will discuss the current status of data science in the game industry. I will review the main techniques used to predict player behavior, focusing on learning methods that serve to reduce churn and highlighting some of their appealing features: in particular, they are effective in real business environments, do not require previous manipulation of the data, can easily adapt to various types of games and players –that is, to different data distributions– and deal efficiently with the temporal dimension of the churn-prediction problem.
The IEEE 2017 Conference on Computational Intelligence in Games hosted a game data mining competition, in which teams were to accurately predict churn within NCSOFT’s game Blade and Soul. In this talk, I will also present the feature engineering and modeling techniques that led us to win both tracks of that contest.
Dr. África Periáñez is the Chief Data Scientist at Silicon Studio and also the Head of Yokozuna Data (yokozunadata.com). She leads a team of data scientists and engineers who develop a state-of-the-art machine learning platform that predicts individual player behavior. She is a senior data scientist and researcher with more than 12 years of experience. África holds a PhD in Mathematics from the University of Reading (UK), a Master of Advanced Studies in String Theory Physics carried out at CERN and an MSc in Theoretical Physics from the Autonomous University of Madrid.
She has worked as a research scientist in renowned institutions such as CERN (under a Marie Curie Individual Fellowship), RIKEN in Japan (working with the world’s fourth-fastest K-computer), DWD in Germany (German Weather Service, working in Satellite Data Assimilation) or the University of Reading in the UK. She is the co-author of multiple peer-reviewed articles and has been a speaker in more than 20 international conferences and many regular seminars. She also has significant industry experience at SPSS Inc., IBM and now Silicon Studio, and her work has been featured in various publications, including Bloomberg, Sinc, The Japan Times, The Washington Post and The Economic Times.
África has a goal in mind: to democratize Game Data Science (so that it can be used in operational business settings) and push the frontiers of this field to new heights. Using cutting-edge machine-learning and statistical techniques, she aims to obtain a thorough understanding of players’ behavior and to accurately predict their actions in order to contribute to the development of amazing data-driven games. Her main research interests include ensemble-based methods, deep learning applied to time-series forecasting and Bayesian approaches to predict player reactions.