Computer games have been front-runners in many important developments in the IT industry, including digital distribution, cloud storage, user driven design, and crowd sourcing. So it’s not surprising that game developers are in a leading position when it comes to big data analytics and machine learning.
In the vastly sophisticated and competitive world of today’s online games – which generate revenues that rival Hollywood – the drive to create a unique user experience has driven innovations in machine learning and big data analytics that outstrip the advances in all but a handful of other industries. Online games have the ability to monitor all aspects of player behavior, so, just as Google is able to refine your search results by analyzing your previous searches and comparing them to the billions of searches done every day, online game companies are able to modify game behavior to ensure a more optimal game experience by observing what works – and what doesn’t – in the gamer’s world.
For example, Riot Game’s League of Legends is arguably the most played game in the world, boasting over 70 million registered users and 12 million daily active users, and more than 1 billion hours played per month. This much player interaction creates massive amounts of data, and Riot wastes none of it.
League of Legends allows players to join teams of “champions” that compete in combat. Champions are pre-defined fantasy characters with varying talents – magic, sword fighting, resilience, and so on. The success of the game depends on champions being roughly balanced, so that there is no overwhelming disadvantage for a particular player. To ensure balance the online software monitors and adjusts champion performance to ensure that no particular champion is over- or under-powered. Similar learning algorithms are employed to provide “matchmaking” services that ensure that teams are constructed from balanced groups of players – allowing novices and gurus alike to enjoy challenging game play.
League of Legends is free to play, but players typically pay real money to acquire advanced champions, weapons and other virtual artefacts. The existence of a virtual economy in most online games generates challenges for game optimization and analytics. For instance, the game Guild Wars employs a hybrid market, partly controlled by player interaction, and partly managed by the game itself. This hybrid design helps to avoid the bloating of prices and market crashes, while allowing a realistic simulation of a real market. Predictive analytics on these virtual market trends can be used to predict player behavior, including future growth of player populations in various “professions.”
The science of economics is famously hampered by the inability to experiment with real-world economies. In the virtual world, however, experiments with economies can happen rapidly and with little real world consequence It’s likely that research on these virtual economies will lead to a better understanding of the dynamics of real world economies, and, the same economic analyses pioneered in Guild Wars could be used to predict how the job market will look in future years, as well as to more accurately predict market bubbles and collapses.
Sociologists and psychologists are also using data collected in online games to examine how people form groups, compete, co-operate and interact. With every action of millions of real players available for study, difficult questions in these fields are becoming easier to answer.
The analysis of the data generated by online games is essential to improving the user experience and competitiveness of these games. But, this “big data exhaust” has a broader applicability – the data generated from these virtual worlds can tell us important things about the real world.
About the authors:
Guy Harrison (guy_harrison@dell.com ) is an executive director of R&D at Dell. Michael Harrison (michael.j.harrison@outlook.com) is a undergraduate computer science student at Monash University.