UBC-O offers an award it calls the Undergraduate Research Award, granted to a small number of students that apply, and they get a living wage($6500 this year), and a budget($1500 this year) for supplies. Seeing as I am in Computer Science, I don’t really need a budget, just access to powerful computers.

What I want to do is quite interesting to me, and a faculty member has expressed interest in the topic. First, consider online multiplayer games. There is a subset of users, growing, that can be called griefers. They purposefully ruin the enjoyment of the game for others, either in the game, or out. Griefers are a serious problem for people that want to have fun and they are a considerable source of ‘churn’ in online games.

The problem is how do you recognize the actions of a griefer? Is there a way to recognize a temp-griefer, someone that griefs other players rarely? Griefing guilds? One solution is human judgement, but that itself is prone to various pressures, like cost, beliefs of the human employees, etc. What I want to look at is if there is any way from a basic corpus of data, like wins/losses, rankings of the match/opponent, friendships, playtimes, etc, and determine algorithmically whether someone is a griefer. This algorithm would have to take into account player immaturity, bad losers, anger, and fraud, which all biases the data to essentially “lie”.

I’ve been quantifying various aspects of one-on-one games, such as maturity, strategy, et al that are game and platform independent. An interesting problem I ran into is that without a psychological test, the only basis for determining maturity is against other player’s maturity. Say we had player Alice’s responses to her victories plotted as a historigram, with time as the x-axis. Then, if we plot her opponent’s reactions to their loss, we see that if Alice is a gracious winner, ie, high maturity, then that is reflected in her opponent’s responses.

However, the opponent’s responses are skewed with respect to their own maturity… and you can see how it becomes an infinite recursion.

So thats one avenue of attack, as this recursive structure is amazingly similar in structure to the problem PageRank solves. Another avenue is to use known maturities, ie staff members, however,  not every such network will have a path from a staff member to an unknown player.

If you’ll notice, that my plan is to avoid using any population data, such as age, gender, race, etc, as there will always be outliers, and instead use past performance and reactions, to determine relatively quickly a new player’s intentions.

Imagine if you had a network of players with known high maturity level, then many measures of stemming and even discouraging griefer’s become obvious, that work on a purely social level. One behaviour of griefers is to create new throwaway accounts, and if we restrict these new accounts only to playing against mature, skilled, encouraging players, then if said players identify the new players as being a detriment to the community, ie, griefers, then it becomes easy to deal with those new players very very quickly before they actually cause a problem for the community.

What do you guys think?