Feb 14, 2019

How Data Science and Machine Learning can help create better games

, , ,

What makes a game fun to play? At Massive, we not only develop games – we also research them and use different innovative technologies and tools to understand the motivations behind playing them.

In this article, Alessandro Canossa, who works as a Data Scientist in our Massive Consumer Experience team, talks about the research he and some colleagues have done on player motivation in Tom Clancy’s The Division and how it can help shape gameplay as we know it.

Every developer wants to make a game that is fun, but fun isn’t the same for everyone.

When I joined Massive and Ubisoft two years ago, I was thrilled to work at a company where I could actively work with the exciting challenge of deducing advanced intelligence from gameplay data.

Before I started working at Massive, I was Associate Professor at Northeastern University in Boston working with the burgeoning field of Game Analytics, and I even wrote one of the first books on the subject.

Thanks to my academic past, I had the opportunity to work together with my previous colleagues David Melhart, Antonios Liapis, and Georgios Yannakakis at the Institute of Digital Games at the University of Malta on a special project: predicting what motivates different people to play a game based solely on their gameplay behavior.

So which basic psychological needs are satisfied by different video games? What makes a game actually fun to play?

To answer these questions, Massive Entertainment has developed a questionnaire called “Ubisoft Perceived Experience Questionnaire” (UPEQ). The questionnaire is based on a model called Self-Determination Theory, and reveals how each individual player is driven by a different configuration of these factors: competence, autonomy, relatedness and presence.

However, no matter how good a questionnaire is, it is difficult to collect responses for the millions of players engaging with a game.

But, by harnessing innovative machine learning techniques, we found a way to predict player motivations based solely on gameplay behavior and the questionnaire responses from a handful of players, at almost certainty levels. The details of this study can be found here.

This is an incredible result in itself, because normally we would only have information on motivation from the few hundreds of players who take a survey. Now, we can infer it for all the millions of players that play our games.

To predict player motivation the researchers used data gathered from almost 300 voluntary players of Tom Clancy’s The Division and ran it through a preference learning algorithm.

Using game metrics, playstyles, and the UPEQ, the algorithm was ‘taught’ to predict presence and the motivations of players as defined by the psychological theory of self-determination, namely competence, autonomy, and relatedness.

After training, the algorithm can predict how a player feels in relation to competence, autonomy, relatedness and presence with up to 97% accuracy based only on the gameplay metrics.

It is no surprise that motivations and desires affect how someone plays a game, but tying specific in-game actions to specific motivational factors has proven tricky.

Preference learning and solely measuring the relations between the data points is a crucial innovation leading to a better understanding of player behavior and the accompanying motivations.

Player experience is subjective and because of this, its measurement seems to be relative.

The gameplay of players demonstrating a high degree of competence differs greatly from those who lack competence, but there is a whole array in between where the distinctions are not so obvious to the human eye.

The ordinal machine learning approach, however, is able to predict complex psychological constructs in all of its subtleties and nuances. And the success of this experiment shows that it is possible to use high-level gameplay metrics and survey-based annotation to model complex emotional and cognitive states.

At Ubisoft we will leverage this deeper knowledge about our consumers to provide better, more personalized gameplay experiences in the future.

  • Are you interested in leveraging innovative AI techniques to push Ubisoft games to the next level? We now have open positions for Data Scientist, Lead Data Scientist, and Junior Analyst.
  • Want to learn more about about the “Ubisoft Perceived Experience Questionnaire” (UPEQ)? Read about its creation and validation here.
  • If you want to read up on the subjective nature of player experiences, check out the article here.

More to read

Discover more articles