Classifying Players For Unique Game Experiences

togelius writes "Whenever you play a game of Tomb Raider: Underworld, heaps of data about your playing style is collected at Eidos' servers. Researchers at the Center for Computer Games Research have now mined this data to identify the different types of player behavior (PDF). Using self-organizing neural networks, they classified players as either Veterans, Solvers, Pacifists or Runners. It turns out people play the game for very different reasons and focus on different parts of the game, but almost everyone falls into one of these categories. These neural networks can now quickly determine which of these groups you belong to based on just seeing you play. In the near future, such networks will be used to adapt games like Tomb Raider while they are played (e.g. by removing or adding puzzles and enemies), so you get the game you want."

Comments

Re:Foruc on different parts of game

[ShieldW0lf]

Just what we need... surround ourselves with ourselves. That will challenge us and cause us to grow into intelligent, tolerant and well rounded individuals.

Re:Almost everyone?

[gnick]

But, once you've run your data through and decided that 4 categories are sufficient, most designers (including myself) will restrict the NN to those categories. And somebody with really weird behavior will get lumped in and will slightly skew the existing category. The guy who runs into a crowd and dies over and over again may be described as a Runner, but he'll be an outlier in the runner class and his behavior will tweak the definition of a Runner.

Your options are to ignore outliers like him to avoid polluting your class, add a new class for people with that kind of behavior if there are enough of them to justify it, or (most likely) just accept that outliers skew tight groups and lump him in as a Runner - If the group is tight enough and he's rare enough, it won't matter.

Ideally, however, your architecture will be flexible enough that you can weigh how good a fit each player is to each group and adjust accordingly. I.e. adjust every obstacle according to a best-fit weighting rather than just delivering 4 different options on each level. Not having played the game or reading TFA, I can't speculate on that front.