This is how the developers of The Division 2 use data science and machine learning to create better games.
In a blog post by Massive Entertainment, the developers of The Division and The Division 2, Senior User Researcher and Data Scientist Alessandro Canossa poses the question, “What makes a game enjoyable?”
Because at Massive, as Canossa explains, they do not just develop games. They also study them and use various innovative technologies and tools to understand what motivates players to play the games.
“Fun” is not the same for everyone
What motivates players? Canossa explains that every developer wants to create a game that is fun. But fun is not the same for everyone. That is why he is working on a project that aims to predict what motivates players in a game solely based on their gameplay behavior.
In other words: He is examining which psychological needs are satisfied by different games and what makes a game enjoyable.
Who is working on the project? Canossa himself was an associate professor at the university in Boston, working in the growing field of game analytics. Due to his academic background, he was able to collaborate with former colleagues at the Institute for Digital Games at the University of Malta to advance this project.
Project aims to collect data from millions of players
Questionnaires collect only limited data: To explore what makes a game enjoyable for gamers, Massive initially developed a specific questionnaire. This is based on the self-determination theory and outlines how much players are driven by the following factors: competence, autonomy, social relatedness, and attitude.
But no matter how good a questionnaire might be, it is difficult to gather comprehensive feedback from millions of players.
Innovative technologies instead of questionnaires: Therefore, as Canossa states, they are now using innovative machines that learn certain techniques to predict player motivation based solely on gameplay behavior.
This technological advancement alone is already incredible, but now they can evaluate not only the motivation data of a few hundred players who filled out the questionnaire but also make use of the data from millions of players out there for future games.
This is how player motivation can be predicted
Machines must first learn: They used data collected from around 300 volunteer The Division players. This data was processed through a specific learning algorithm.
By preparing gameplay data, play styles, and the questionnaire, the algorithm was continuously taught how to measure players’ commitment and motivation.
After the learning phase, the algorithm can predict how a player feels—depending on their competence, autonomy, social relatedness, and attitude. Based solely on gameplay data, the algorithm is said to have an accuracy of 97%.
The gameplay of players who exhibit high competence differs significantly from players who are not as competent. This can also be observed with the naked eye. However, there are many gradations in between that are not so obvious to detect.
And it has proven tricky to determine motivational factors based solely on in-game actions.
The technology can do it: The machine, however, is now capable of predicting complex psychological constructs in all their nuances based on gameplay.
The success of this project shows that it is possible to use gameplay data, connect it with survey-based insights, and thus represent complex emotional and cognitive states.
Better games become possible: At Ubisoft, they will leverage this deeper knowledge about consumers to deliver better and more personalized gameplay experiences in the future.
The new clan system in The Division 2 may motivate you to achieve goals together with your friends.
