Bayesian Analysis of Online Game Data
|
|
Bayesian Analysis of Online Game Data |
|
Ralf Herbrich, Thore Graepel, and Tom Minka
Online games, in particular the Xbox Live service, provide a wealth of information generated by the participating players as well a many challenges for modelling and inference related to ranking, matchmaking, cheating, tactics etc. We use probability to tackle some of these tasks.
Ranking/rating players in online games according to their playing strength is essential
for good matchmaking and can provide additional incentives to players via leader
boards. In this project we apply probability theory to estimate the playing
strength of players from game outcomes. In particular, it is essential to
quantify the uncertainty in the estimates and to be able to deal with multiple
players / teams, where the resulting game outcomes are rankings among the
participating players including draws.
Virtual online gaming communities are not unlike other human
communities in that there people who cheat or disturb the functioning of the
community in other ways. Online services such as Xbox Live provide players with
an opportunity to leave feedback on other players, e.g., accusing them of
cheating or complaining about the use of abusive language. In this project we
build models that make it possible to infer who are perpetrators and who are
reliable sources of information from the feedback data. We employ probabilistic
modelling techniques to deal with noise and uncertainty associated with the
data.
There is no publication available about this work at the moment, but please look at the TrueSkill™ pages (link below).
Machine Learning and Perception—Machine Learning—Bayesian Analysis of Game Data