Cheaters in the Steam Community Gaming Social Network

I found this paper in Technology Review some weeks ago and, since I've been working with gaming data, I decided to read it. Although being a pre-print version, this work has attracted a lot of interest recently. Researchers from University of South Florida, University of Delaware, and University of British Columbia collaborated in this paper that studies social interaction in the Steam Community, which is the most important gaming community in the world. In particular, the work characterizes several aspects of cheaters (i.e., people that violate the rules games using software) through social network analysis techniques. I must confess that this is not an easy paper to summarize.

The study is based on a crawled dataset composed profiles of 30M users that interact through a social network. A second dataset, containing data about 10K players, was collected from a particular server. Users are classified as cheaters or non-cheaters in a given game by an anti-cheat system provided by Steam.  More than 720K users from the dataset are classified as cheaters, who are blocked by the majority of the available servers.

Cheaters were found to present several interesting characteristics that distinguish them from the other gamers. They own less games, play less, are more interested in multi-player games, and use more restrict privacy settings. Moreover, cheaters are likely to be friends with others cheaters (homophily) and loose more friends than non-cheaters. On the other hand, cheaters and non-cheaters have similar numbers of friends. The results show also a strong correlation between friendship and in-game interaction. Cheaters interact less with the remaining of the community, indicating a possible reaction against their behavior. Regarding group relationships, the authors found that cheaters are not equally distributed among groups, which is an evidence that cheaters learn about cheating on groups. A geographic analysis of cheaters and non-cheaters shows that both playing and cheating are not well distributed among countries. The authors argue that cultural aspects may be related to cheating behavior, something already found by other studies. Another interesting result presented in the paper is that cheating propagates over time, which means that new cheaters are likely to be friends with previously discovered cheaters.

This is one of those papers many relevant findings but few technical contributions. It was hard to read due to the number of analysis made and the brevity of the discussions. I would like to see more general conclusions instead of punctual findings but I confess that I'm not an expert on online gaming. The authors provide an extensive coverage of the online gaming literature and also seem to be familiar with online gaming concepts in general. I hope to see a new paper combining the discoveries of this paper and machine learning techniques in the identification of cheaters.

Link: http://arxiv.org/abs/1112.4915