BZFlag Social Network Analysis

[John Paolillo]

I have conducted a longitudinal Social Network Analysis of BZFlag public server data (as acquired by strayer, http://bzstats.strayer.de/ ), for the period from October 28, 2006 (heyday of Boxy Wars) to Nov 11 2007. My main goal in the analysis was to learn about community structure, and its maintenance in BZFlag, game playing being an important part of that. If people are interested in viewing the results, they can be found here:

Social network over time (movie, different formats, same content):
http://ella.slis.indiana.edu/~paolillo/ ... 1.1.pdf.gz (gzipped Adobe PDF, 73 MB)
http://ella.slis.indiana.edu/~paolillo/ ... .1.mov.zip (Quicktime, zipped; works for me, but ymmv)

I apologize in advance for the inadequacy of the formats. I have a smaller quicktime movie which is not uploading correctly. I still have substantial work to do on making these graphics more legible and controllable, but what is here is sufficient to learn some interesting things. If you're the thoughtful type, you may want to download the pdf; it's the original format of the graphics, and with a little arithmetic, you can figure out which pages represent which days.

A typical frame looks like this:



The entire series of graphics/movie shows the dynamics of the social network as it develops over the >1year time span from Oct 28 2006 to Nov 11 2008. Each frame is a one-day snapshot of the network, and the actors in the network change for each day. The layout is based on an analysis of player-server sessions by five-minute time-slots during the day. These time-slots were clustered, effectively into time-ranges, where red begins at midnight UTC, increasing clockwise, and blue-green represents approximately 12:00 (noon) UTC. Players are represented by points, colored according to the timeslot that they are most closely associated with. The positions of player points show strongly-connected players toward the center while leaving a hole in the middle of the diagram for readability. Links indicate shared player-player sessions aggregated over the entire day, where players share a minimum of 45 minutes together on some combination of servers. Weight of the link is proportional to the time players spent together. On any given day there are around 2000 players online; all of them are represented in the plots.

The most striking things to me are (1) weekly patterns in player volume/density of connection, wherein weekends have more players and stronger inter-connection among players (2) the clusters of users around mid-morning GMT (greens) and afternoon GMT/EDT/PDT (cyan to magenta). I'm somewhat time-zone challenged, so I've been going back and forth on my interpretation of this a bit, but they appear to represent different playing (sub-) communities, possibly Australia/Asia and European/(North?) America. The two are typically connected, although on many occasions that connection weakens considerably, and for sustained periods of time.

Another analysis shows that the patterns in connectivity are strongly periodic, with strong harmonics (like a plucked guitar string). The daily and weekly cycles are obvious, but longer and shorter time cycles show up as well.

John Paolillo

~edit: link fixed; gzip posted in the interest of saving bandwidth~

[joevano]

I am interested in looking at the pdf, but the link is dead (404, for me)

[John Paolillo]

Sorry for the link/format faux-pas. It should work now.

John

[quantum dot]

hmmm,

this is getting more interesting. I have worked on complex social networks myself and I might have something to add to this preliminary analysis of yours. You have chosen to link players according to their playing times, however, in my opinion, the link would not be an actual social link unless they have played the very same servers during that time slot. Some people play at the very same time I do, who I have never seen around.

There must be much more information to be uncovered if you link two players who played at the same server at whatever time slot of the day, later you can refine your algorithm to distinguish those players who play on the same sever but they rarely see each other because of the time region they live in. So the actual community analysis should in my opinion be performed the other way around: first servers, then times.

This would give the real community structure of BZflag. I would say one should clearly see a community structure emerging in this analysis because most players only play a few servers, maybe 4 at most. They really dislike or feel little to none attraction to other servers. For doing this you would need some internal knowledge of bzflag styles, for instance to group ALL GU League servers in one single server, same for 1vs1 league or ducati league.

It would also be interesting to add centrality and betweeness in this analysis. My simplest guest is that some admins are in the center of these really popular servers. Someone like "menotume" for instance should pop out as a playing a central role in holding the community tight.

On a side note, the number of player from Asia, Australia or South America is very low, extremely low I would say. The reason being that the immense majority of servers are in USA and EU. Therefore, the distance from Asia to the popular servers gives prohibitively high lag to connect and play. Most players are from USA and EU (maybe as much as 95%), with a little community from Brasil and Argentine. Little can be as little as 5-10 of regular players in total. There are no Asians or Ausies that I know of who are currently active (in the popular servers) and I have been here for 5 years.

[Green Manalishi]

On a side note, the number of player from Asia, Australia or South America is very low, extremely low I would say. The reason being that the immense majority of servers are in USA and EU. Therefore, the distance from Asia to the popular servers gives prohibitively high lag to connect and play. Most players are from USA and EU (maybe as much as 95%), with a little community from Brasil and Argentine. Little can be as little as 5-10 of regular players in total. There are no Asians or Ausies that I know of who are currently active (in the popular servers) and I have been here for 5 years.

on badgerking's (a server in the central usa), there are at least 5 regulars from australia and about 4-5 regulars from eastern asia. only one player from asia seems to have consistent lag problems.. on the other hand, to the best of my knowledge, there are only very few regulars from south america. actually i only know one guy from argentine.. my observation is based mostly on a single server, but i'm still surprised you think there are no (or hardly any) active players from australia and asia.

[John Paolillo]

I'm back from the Tampa Bay area, where I presented this research, and finally had a chance to post my slides. They can be found at the following URL:

http://ella.slis.indiana.edu/~paolillo/ ... unbelt.ppt

Note that the slides link to a Quicktime movie, which is just the social networks movie in the opening post. There is not a lot in the way of prose explanation on the slides, so if you think you'd like some, you may have to wait until I can get a written version of the talk. Once I clear some prior things off of my desk, I intend to take this up again in earnest, looking for a place to publish the results. I'm grateful for the interest in this research in the forums and would welcome any feedback players might have. Research on the other fronts (BZFlag in-game chat, use of space) goes on, and some graduate students are joining the effort as well. I'll post updates as they come up.

Best, and happy tanking,

John Paolillo

[too much loving]

Thank you for sharing the slides. It is good fun to watch from the side line

Anyway, it is also fun to play the devils advocate, so I wondered what would happen if you drew a "social network graph" for a single person game like a web based mine sweeper. You could do that by counting how many players happen to play simultaneously on a given time of day every week. That would result in a "social network graph" like the one you posted in the begining of this thread. My question is whether the "minesweeper social network graph" would look radically different from the "bzflag social network graph"? Perhaps you could distinguish between the two games by testing for something like the following.

* I tend to play for a longer time if there are many players on my favorite server.
* I tend to play for a longer time if a new map becomes popular.
* I tend to play for a longer time if my opponents have good connections.
* I tend to play for a longer time if a given other player is also playing.
* I tend to play for a longer time if I get a high score or if I have skillful opponents.
* I tend to play for a longer time if people respond to my wonderful jokes.

This would prove that the choices of one bzflag player depends on the choices of others in a way that is different from the "single player minesweeper community". (It probably does, I would just like to see the graphs)

By the way if you have a video of a talk then I would love you to post it somewhere. That would make some of the graphs easier to understand.

Keep up the good work