What is meta? ×
Meta Stack Exchange is where users like you discuss bugs, features, and support issues that affect the software powering all 129 Stack Exchange communities.

I've just made a map of all StackExchange sites, except for the 3 biggest ones:

Map of SE sites (but 3 biggest)

(a high(er) res png file) (EDIT: there was a pdf file, but for some reason it was eating >1GB of memory, at least on my computer)

  • The graph is based on users with rep>=200 at a given site.
  • Node size is related to number of such users.
  • There is an edge between two sites if there are common users more often that one would expect on random, see on observed/expected ratio on Stats.SE.
  • Colors are indicating different graph communities (nodes which have much more connections inside a group than with other groups).

On GitHub there is code and a recipe how to make a similar plot (or tweak this one). Also, you may want to look at a Graph Map of Math.SE.

Why not include Server Fault, Super User, and most importantly, Stack Overflow?

SO, to some extent, is a different story from the other sites. When it comes to SF and SU - unfortunately, the SE API (or at least - my Python wrapper, se-api-py) has some issues with obtaining more than 100k items.

Any questions and comments are welcome.

share|improve this question
19  
I love that you keep making these! :D I think they're awesome, and they make me look at the sites in a different way. –  Aarthi Dec 7 '12 at 17:36
2  
I haven't dug into your implementation yet, but I was talking about a similar idea the other day. My idea was to take the text of all of the questions and answers, the tags, and with shared users in order to determine if there were any clusters of related sites. What are your thoughts on this? Do you think I might be able to leverage your work to do this? –  Thomas Owens Dec 7 '12 at 17:38
    
@ThomasOwens In my case I count cases when a user (identified by account_id) has accounts more sites, e.g. ['gaming', 'rpg', 'boardgames'] (each with rep>=200 just to filter out people how are not active at all, otherwise the resulting graph is a mess). Looking at similar texts may be tricky (but good e.g. to suggest migrations), tags at different sites may have different meaning, etc. Anyway, there are many possibilities to mine the SE data. My code is on GitHub, feel invited to use it. When it comes to using my API - I need to clean it up before uploading. –  Piotr Migdal Dec 7 '12 at 17:53
    
I do like your approach of filtering out by reputation. I have an account on several sites, but it's mostly because I can read, comment, and vote. I actively participate on far fewer sites than I have an account on. If I manage to get anywhere, I'll let you know. I'll check out your code when I get some free time. –  Thomas Owens Dec 7 '12 at 17:55
    
@ThomasOwens If you have any questions or doubts about my code - just ask. :) When it comes to setting threshold at 200 - it's kind of arbitrary and perhaps there are better ways (e.g. looking at rep in the last months) or just number of questions posted. –  Piotr Migdal Dec 7 '12 at 17:58
    
Why is ProgSE grouped with Fitness and MartialArts and not the coding group? I thought Programmers code all day long and don't exercise ;-) –  Manishearth Dec 7 '12 at 18:09
    
@Manishearth I guess the links is the following: programing - geekiness - rpg - martial arts. But fitness... - no idea. Or maybe the same thing as with "being productive" vs "reading about productivity" (e.g. the famous The Ultimate Productivity Blog) :P. –  Piotr Migdal Dec 7 '12 at 18:14
    
@Manishearth: Little know secret: programmers like asking each other questions about fitness. Whether or not we actually follow through and do any is a different story... –  Jon Ericson Dec 7 '12 at 18:17
    
@PiotrMigdal could we get a larger image please? –  smcg Dec 7 '12 at 18:24
1  
@smcg I added a pdf - raw.github.com/stared/tag-graph-map-of-stackexchange/master/… –  Piotr Migdal Dec 7 '12 at 18:51
    
I keep wanting to click on all those colorful circles :) There's a Chrome app called "Visual History" that's fun like that. Very interesting graph. –  JoshP Dec 7 '12 at 18:52
    
@Josh Yeah, I want to do it a bit more interactive, most likely with d3js.org, e.g. flowingdata.com/2012/08/02/… :) –  Piotr Migdal Dec 7 '12 at 19:19
    
@PiotrMigdal In your first comment you say you're identifying user by account_id. AFAIK the only field that is comparable across sites is EmailHash. –  Mark Hurd Dec 17 '12 at 0:18
    
@MarkHurd Thanks, I will check that. But do you have any idea why acount_id (AFAIK, an identifier of account across sites) shouldn't work? –  Piotr Migdal Dec 17 '12 at 8:36
    
@PiotrMigdal Yeah, it looks like account_id is an API only thing, and on a bit of testing at api.stackexchange.com/docs/… and api.stackexchange.com/docs/… it does return the same account_id for me (but not id). –  Mark Hurd Dec 17 '12 at 11:03

2 Answers 2

up vote 16 down vote accepted

This graph is interesting to stare at for a while, but the best bit, in my opinion, is the color groups. For the sites I'm most familiar with, the organization makes sense:

  • Hermeneutics fits well with the other religion sites. (Philosophy makes since too, considering how often the concept of God pops up over there.)

  • Travel and Linguistics fit perfectly with the non-English language sites.

  • History and Skeptics fit together as do the Writers-English-SciFi-Movies cluster, though I'm not sure what affinity the two halves of that color have.

  • The Unix sites are associated with the more "hacker"-type sites.

  • Hard science, math, and Acedemia are in the same cluster. (Plus Libraries.)

  • There are a ton of "niche technology" sites. It's interesting that Music ends up in there.

  • The "hands-on" sites (DIY, Gardening, Cooking and so on) cluster nicely.

And then there are the green sites. Are they "Geek Culture" sites? That's my best guess.

share|improve this answer
    
Nice observations! –  Piotr Migdal Dec 7 '12 at 19:34
    
I wonder why the Poker site is in with the language sites. –  Don Kirkby Dec 8 '12 at 4:42
    
@DonKirkby It has very strong ties to chinese, russian and chess (the later not a language, but also tightly connected to languages). So, given the data, the association is not surprising. Why is it, in the first place? I can only speculate... (better to ask people from poker). –  Piotr Migdal Dec 8 '12 at 13:47
    
@Don Kirkby: I had noticed Chess and it made sense because of the association with Russian Language. (Russian speakers have dominated the top levels of the game for a generation or more.) But I hadn't noticed Poker since it's small and tucked to one side. Interesting. –  Jon Ericson Dec 8 '12 at 20:27
    
@JonEricson ...and as a curious observation, parenting is a hands-on hobby. –  Piotr Migdal Apr 24 '13 at 11:31

Looks like the folks over at ExpressionEngine.SE don't get out much.

share|improve this answer
1  
How many 200+ reputation users do they have? It's a very new site. –  Jon Ericson Dec 7 '12 at 18:17
    
It's the newest public beta site. Not that many users, not that much rep. –  Manishearth Dec 7 '12 at 18:19
2  
@JonEricson: 59 right now. Still more than SmugMug's 19. –  Ullallulloo Dec 7 '12 at 18:20
    
@JonEricson (and Manishearth), remember that node size is based on the number of users. Other similar size Stacks look more connected than the larger ones if anything. I think those guys really just don't get out much. –  blahdiblah Dec 7 '12 at 18:42
    
Funny you should say that. The ExpressionEngine proposal had a lot of followers that didn't already participate on an existing site. I'm not sure why, but unlike some of the other programming-related proposals, they didn't just split off of SO and came from the outside instead. –  Anna Lear Dec 7 '12 at 19:29
    
@AnnaLear That is odd. It's almost as if the proposal was promoted by EllisLabs, especially given the short time to launch. However, the only reference I can find leads to the first attempt that fizzled out. –  Tim Post Dec 8 '12 at 2:26
    
@TimPost The proposal was enthusiastically promoted, for example: meta.stackexchange.com/questions/152995/… –  Yannis Dec 8 '12 at 3:38
    
@TimPost I don't think EllisLabs promoted, but it certainly got some unaffiliated publicity. The specifics escape me at the moment, but it was really cool to see at the time, since growing a community external to our network is one of our bigger challenges right now. –  Anna Lear Dec 8 '12 at 4:01

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .