« Grouping and Guild Recruitment | Main | Guild Members: Time Spent Together »
October 07, 2005
Mapping Social Networks
Server Sample: RP (High), PvE (Medium), PvE (High), PvP (High), PvP (High)
Sampling Period: 8/01/2005 12:00 am - 8/30/2005 12:00 am
Sampling Resolution: ~12 minutes
Parsing Method: The sample unit is each unique character. Each character was tracked across the server logs. Total playing time, lowest observed level, highest observed level, guild affiliation, and zones seen in were parsed.
Data Filter: None
Sample Size: 241,378 characters
We've been working on social network data at the guild level and would like to give an overview of our approach before getting into the gory details of what we've found. As many of you already know, there are only a few variables that we are able to get at from the client-side, so actual character interaction is something we must approximate via proxy metrics. We'd like to describe the metrics that we have used, but please do not hesitate to suggest others that would be possible with the set of variables the client-side currently has access to.
Guild Roster: Over the sampling period, we generated a list of all guilds we observed. Then we generated a roster for every guild consisting of every character who has been observed to have that guild tag.
Co-Presence Metric: In each snapshot, find all members of each guild that are online. For each observed pair, increment the connection weight between these two characters by 1. In other words, this metric tabulates overall guild co-presence - the frequency at which members of a guild are online at the same time.
Co-Location Metric: In each snapshot, find all members of each guild that are online and in the same zone (and only if the zone is not a main city zone). For each observed pair, increment the connection weight between these two characters by 1. In other words, this metric serves as a proxy for collaboration - the frequency at which members are working together at the same time.
We've found that the co-location metric provides more readable and comprehensible graphs and have used this metric for most of our analyses. Below, we present some social networks of guilds constructed using the co-location metric. We have removed the names of characters but left in their class and level information.
In the following social network graphs, connection weights are based on the co-location metric (with a minor threshold applied to exclude very weak ties). The weight of each line implies the collaboration frequency over the sampling period. Each node is marked by the character's class and level progression over the sampling period. The more connections a node has, the more central it is placed in the graph itself. Characters who were never co-located with others during the sampling period are depicted as free-floating nodes. All 4 guilds depicted had between 40-50 members. Therefore, the connection weights themselves are directly comparable between the guilds.
The social network in these 4 guilds are somewhat different. In the top-left example is a fairly low-level guild. Most of the members are in the mid 20's. It appears that most guild members do not play with each other. In the top-right example is a mostly high-level guild where most members do work together. It is easy to pick out the pairs that seem to work together the most (such as the mid-level druid and mage).
In the bottom-left example, we have a guild where we see more distinct cliques. There's a somewhat hard-core 4 character cluster on the left-hand side, a mid-level triad on the bottom left, a mid 20's clique that's held together by the druid in the middle, and finally a more casual low-level clique on the top right.
 |
 |
 |
 |
Posted at October 7, 2005 11:36 AM
Trackback Pings
TrackBack URL for this entry:
http://blogs.parc.com/cgi-bin/mt-tb.cgi/48
Listed below are links to weblogs that reference Mapping Social Networks:
» Mapping Social Networks in WoW from Quantum Gaming
The folks at PlayOn have a fascinating post up on their work mapping social networks within World of Warcraft at the guild level.
The data and methodology are interesting, but what's really cool are the visualizations of the data. The method of visu... [Read More]
Tracked on October 7, 2005 05:14 PM
» Tracking Interaction from reBang weblog
Nick Yee posted on Terra Nova that he's updated the PlayOn blog with an entry showing social network maps derived from "World of Warcraft" data they've been collecting. I'm curious to know how this aligns with real life organizational social maps ... [Read More]
Tracked on October 8, 2005 10:24 AM
Comments
This is a really incredible resource.
For example, using this data, you could quickly determine the method guilds use to expand their membership. Do they recruit high-level characters? Train up low-level characters? Bring on low-level characters and let them flounder?
You could determine which classes gain levels fastest - and which classes are best at helping people gain levels. You could determine whether people tend to play in-guild a lot, or whether they have friends outside the guild they tend to group with, or whether they solo, or whether they band up with random people. And what classes tend to do which.
And so much more! You could determine how many mules there are, how many farmers... this is so useful!
Posted by: Craig at October 7, 2005 02:36 PM
This is fascinating. The designer in me is geeking out a bit: it would be great to see color introduced to represent additional data points without overly cluttering the visual. You could also use circumference to indicate level, as opposed to numbers. Shape could be used to represent class.
Nice work. Very cool!
Posted by: Dave C. at October 7, 2005 05:12 PM
It'd be interesting to see this data also linked to the playtimes of each player. Do people who play 12+ hours a week group more than their more casual guildmates? Or do they solo more?
Posted by: Devin at October 7, 2005 05:27 PM
Dave: I am working on a little visualization tool that will do just what you suggest - and more :) Stay tuned.
Posted by: Nicolas Ducheneaut at October 8, 2005 02:35 AM
This reminds me of charts I've seen used to determine manufacturing efficiency on assembly lines; mapping out interactions among workers within a cell, for example. Or something of that nature.
It'd be interesting to see player character equivalents assigned (a mid-level mage might be a mid-level R&D researcher in a RW corporation) and then determine if they reasonable match RW interactions. I get the feeling they'd look pretty similar.
Posted by: csven at October 8, 2005 10:18 AM
Nicolas: Excellent! I look forward to seeing the results. Let me know if I can be of assistence - I was a graphic designer and Flash programmer in a former life and love to collaborate. :)
Posted by: Dave C. at October 8, 2005 12:19 PM
Nick,
I've been following your work for years. We frequently refer to it and cite you on a game development forum. When quoting an article you wrote in Daedalus to Ryan LaSalle, I just found out that you used to work at a certain Tech Labs that I'm associated with... and that filled me with pride. :)
Add that as one virtual social link to your growing database...
Posted by: Jay Crossler at October 12, 2005 01:12 PM
Speaking of interesting, but probably less straightforward cross-correlation studies, I wonder how the metrics you earlier used for looking at player motivations would map onto the players of the characters whose interactions you tracked here? Alas, I suppose that would require specifically recruiting study participants, which might introduce bias, and certainly would take substantial resources.
Posted by: Evangolis at October 13, 2005 03:00 PM
Hey Jay - small world isn't it :)
Nick
Posted by: Nick Yee at October 14, 2005 03:09 PM
I don't suppose there's any available software that would would make it possible for me to create one of these diagrams for my own guild?
Posted by: Bill at December 14, 2005 07:24 AM
Bill, we're working on such software and we *may* release a version for guild leaders at some point. The problem is getting the data to feed to the tool - you'd need to deploy a bot on your server to monitor your players' activities 24/7. This would require getting an extra account.
Posted by: Nicolas Ducheneaut at December 14, 2005 01:53 PM
Nicolas, if you do release such a tool, consider using a distributed approach. There's no need for one client to be collecting all the data, since you can use server time as a strong identifier of a sample point, you can have multiple clients collecting data to provide a greater degree of coverage.
Posted by: Byron Ellacott at February 6, 2006 04:10 PM
Just take a page from the CensusPlus mod. You don't need one client gathering all the data for everyone when you can have every client gathering the data for it then ... you know, I think a pretty simple mod for CensusPlus would get you the data you needed. You'd just have to map it out somehow. I have no idea how you'd go about making one of those fancy graphs that Nick Yee used up there.
Posted by: Bart Humphries at February 13, 2006 02:04 AM
Wouldn't the co-presence metric observed over a long time (say a year) allow to find out how many characters are alt characters? At least the non-co-presence would be a higher bound not to far from the real percentage.
A faster method to find alts might be to count all characters in a guild present in a short time frame (say 30 to 60 minutes) but never co-present. This would count players who switch to their alts for short house-keeping operations (like reading mail, AH...), which is something commonly done with alts. But you would need a high sampling frequency, under 2 minutes.
Posted by: Holger Hellmuth at May 4, 2006 07:18 AM
how to help poor people?
Posted by: Himal at October 11, 2006 09:14 PM
Fascinating data and I'd love to see it both for my old guild where I would have been an outlier and my new one where I'm an officer being asked to take over as GM.
I wonder if you could get some picture of out-of-guild social networks through frequent co-location and grouping data. I'm not sure if there is a way to tell who people are grouped with through.
Posted by: Tom at November 1, 2006 12:57 PM