Meme: LJ MindMap

[chanaleh]

(I'm not sure I quite get this. Okay, it looks pretty and vaguely interesting, but what does it mean? I already looked this and this over, and don't feel particularly enlightened.)

sponsor: photo puzzle

MindMap

Comments

[ailsaek.livejournal.com]

Yours looks a lot like mine. I can't figure out what mine is supposed to mean either, especially since my map had only one person from my friends list on it.

[clauclauclaudia]

Basically, it identifies a core or tier of people who are mutually friended, and then others who aren't all mutually connected, sort of measuring how tightly-woven your social web is.

So, not only do you have mutual links with obra, and jbsegal, and exponentialdk, and onetwolittleb, avacon, oonh, tcb, crs, and chrysaphi... they also all have mutual links with each other. Which when you think about it, really is a neat little bit of information to extract. Then the next tier is cthulhia, rjpb, and tenore, who have mutual links with you and most but not all of your "core" tier. And so on.

Apparently I don't share enough mutually interwoven links to even appear on your mindmap, which is sort of odd considering how much of our friends lists we share. But it doesn't so much measure that as those networks in which everyone has everyone else friended. I only appear on the fringes of my own!

I thought it was cool enough that I made a donation and asked for a color one.

[sweetmmeblue.livejournal.com]

I wonder how tafkar winds up on your map

[bethr.livejournal.com]

Weirdness: I'm not on your map; nor, I think, were you on mine. I wonder what that means?

No, I Take That Back

[bethr.livejournal.com]

You're all squoozed up in one of my corners.

[mrmorse.livejournal.com]

I went reading up on some of what mcfnord has to say about it, and it looks like he's being deliberately obscure. The obvious stuff is that it analyzes your Friends list (and your Friends' Friends lists) and finds groups of people who all list each other as Friends. The larger the group and the more tightly linked (people in a particular group Friending other members of the same group), the more significant the group is.

He then takes this information and applies size, color, and location to indicate the groups. As far as I could find, he hasn't specified how he sets each characteristic, but it appears that closer and larger names are more connected to you than smaller and farther names.

The calculations to generate the map are very computationally intensive and become moreso as the number of Friends you have increases, to the extent that sufficiently large numbers of Friends can break the system. This may mean that Friends of yours who in turn have a high number of Friends may not appear correctly on your map. (Since mcfnord is so vague, I'm not sure that's happening, but it may be.)

The two useful pieces of data you can get from the map are that people who are close to you have a large number of Friends in common with you and Friends who are clustered on your map are likely to share each other as Friends. Without a more concrete explanation of the significance of size and color, it's difficult to say more.

[mcfnord.livejournal.com]

I don't answer every question. Everyone wants concrete answers. Nobody wants to take on the idea I propose in my weblogsociology essay, that the answers are in you, and not me.

The big names in the middle are not just closely linked to the "seed" (the name in the center), but to Each Other as well. That's the Secret Sauce of the LJMindMap.

The computational complexity claudia_ and others experience is not so much about the # of friends she has, but about how interconnected they all are with each other. My post in lj_research speaks to this computational complexity. I'm very sad that some of the super-networkers who have the most interest in this work are inaccessable to it. There is graph theory research on this subject, and I have some half-baked ideas I'll probably try before it's all over.

Size and shade define the first three tiers. In black and white, anyway, the big font bright is tier One, big font dimmer is tier Two, and littler font is tier Three. Beyond that is past three. So if tier One has seven-way friendships, these people are in big and bright. Then probably a six-way or two occupy the same size, just dimmer. Color defeats this dimmer-brighter thing, but it's a tradeoff to get the beautiful colors.

The intrigue derives from the cross-germination of many subtle hints and details. Alone they're just mush.

[mrmorse.livejournal.com]

Because I'm insane, I calculated my LJ MindMap by hand. Don't even think about doing this unless your Friends list is both small and sparsely connected. It took me several hours to complete, and I was only working with 28 people. I'll both go over how I did it and the results I got. If mcfnord is still reading this thread, maybe he can tell me if I did it correctly.

Start with your Friends list. Only include people who have also Friended you. In my case, I had 28 names, starting with:

avon
blkdrgn
chanaleh
cmouse
cnoocy

For each person on your Friends list, go to their info page and look at their list of Friends. For each of your Friends which shows up on that person's list, write down that pair of names. Here's the catch. Both people have to list the other person as a Friend, so the person has to appear in both the Friends list and the Friends Of list. I had 113 pairs, starting with:

avon cnoocy
avon colorwheel
avon cthulhia
avon jadelennox
avon jhango

Now use the pairs to find triples. The rule is that abc is a triple if ab ac and bc are all pairs. What this means is that all three people are Friends of each other in addition to being Friends of you. I have 186 triples, starting with

avon cnoocy colorwheel
avon cnoocy jadelennox
avon cnoocy jhango
avon cnoocy rikchik
avon cnoocy temvald

Note that I've gone from 28 Friends to 113 pairs to 186 triples. This is slow growth. The worst case is on the order of x^n, where x is the number of Friends you have and n is the number of people in the group. If you have 200 Friends, many of whom are also Friends of each other, you could be looking at 10,000 pairs and 500,000 triples. This is the computational complexity mcfnord referred to.

Anyway, repeat the process to get groups of 4. If abc and abd are triples and cd is a pair, then abcd is a group of 4.

Keep going until you've formed the largest groups you can. My largest group is blkdrgn, chanaleh, cmouse, greenlily, mister_nick, scwang, and zenala. Not only are these seven people all my Friends, but they are all also Friends of each of the other ones. These are my Tier 1 Friends and would be in a large, bright font on my map.

I have 17 groups of six interconnected Friends. These include:
avon cnoocy colorwheel jadelennox jhango rikchik
chanaleh greenlily kalessin saxikath scwang zenala
cnoocy colorwheel dougo jhango prog rikchik

Friends who show up here are my Tier 2 Friends and appear in a large, faint font. My Tier 2 friends are avon, cnoocy, colorwheel, dougo, jadelennox, jhango, kalessin, prog, rikchik, saxikath, and temvald.

I have 75 groups of five interconnected Friends. Friends who first appear here are Tier 3 Friends, and so on.

Loosely, my Tier 1 Friends are connected through MITG&SP. My Tier 2 Friends include some people I know from MITG&SP who are less tightly connected and some people I know from around Somerville. I expect that the two groups would show as different colors on a colored map and that the positions of the names would indicate the groups.

I have ordered a LJ Mindmap for myself to confirm my results. Unfortunately, this shows a second computational complexity of the Mindmaps. As interest spreads, demand for maps will grow exponentially, meaning that eventually so many people will request maps that mcfnord will be unable to produce them all. I may have to wait a very long time to get my map.

[mcfnord.livejournal.com]

That's the coolest thing I've read all day!!! Hand-crafted MindMap goodness.

First, growth rate: Exponential? 24 hour server logs tell one story:

   19761 Mar 16
   63144 Mar 17
   28968 Mar 18
  226922 Mar 19
  240417 Mar 20
 2026910 Mar 21
 2160902 Mar 22
 6884692 Mar 23
11283088 Mar 24

It's at least geometric.

But I bet you're dyin' to compare notes... ok, you roped me in... I wanna see if you're right...


Tier 1
mrmorse rikchik jadelennox avon cnoocy colorwheel temvald prog jhango saxikath chanaleh kalessin greenlily scwang zenala cmouse blkdrgn mister_nick

Tier 2
katfairy dragonfriek laurens10 tapuz marfta xiphias evelio

Tier 3
cthulhia dougo jessruth

Tier 4
zogathon

(5, 6, and 7 are empty)



I don't think I met your color expectations... but your first tier is south, second tier north. Neato.

[mrmorse.livejournal.com]

Hmm. I don't get the colors. Maybe they'd become more clear if I studied the source data on some of the other colored MindMaps more closely.

There are some signficant differences between your rankings and mine. Your Tier 1 includes everybody I put in Tiers 1 and 2, with the exception of dougo which I put in Tier 2 but you put in Tier 3. Likewise, your Tier 2 is a mixture of my Tiers 3 and 4, but cthulhia and jessruth are in Tier 4 along with evelio, marfta, and xiphias.

You must be defining Tiers differently than I did. My Tier 1 includes everybody in a mutually connected group of 7 Friends. Tier 2 is everyone in a mutually connected group of 6 Friends, and so forth. That's obviously not what you're doing, but at this point what you're doing isn't clear to me.

I'm curious about the numbers you sent me. Your number of sets at each stage was significantly higher than mine. The algorithm I used ensured that the sets were unordered and unique. That is, if abc is a set, then acb is not also a set. I wonder if that accounts for the difference or if there's something else going on. Now that I think about it, it could be that I excluded myself. (I already know that I'm Friends with everyone on my Friends list, so I could ignore myself.)

Having gone through all this, I have a new appreciation for the impact that large groups of well connected Friends makes. It took me several hours to produce my results, and that was with a maximum set of 7. I've been thinking about it and I think that if my largest group was 8 it would have produced so many more combinations that my work probably would have taken 4 times as long to complete.

I knew I was insane to do it by hand, but I didn't realize how insane. Apparently I'm also lucky.

[mcfnord.livejournal.com]

Your two main groups have enough interrelationship (saxikath in particular) as to have a lot of color crossover.

Your understanding of Tiering is correct. People rise to the highest tier.

I also only track unique sets. Do I exclude the root? I don't remember! If you have a clue on simplifying these calculations... I'm all ears! (Too tired to pay more attention than this right now.) I see a pattern of expansion and then contraction. Indeed, the difference between a "normal" calculation and a "hard" one is like 5-10minutes vs. 5-10days or weeks. There seems to be little in-between.

[mcfnord.livejournal.com]

Your message is still on my mind. How did your numbers significantly differ from the ones I emailed you? I believe I'm doing this without duplication of sets. I believe I have just run up against complexity... if a group ultimately has eight members, then there's about seven times seven or 49 groups representing this fact in the previous step. And for each of these 49, there are about 6*6 or 36 sets in teh step before that. So finding the 8-way goes through this misery of 1764 emerging sets just two steps prior.

I am not a skilled set theoreitician. The long and short of this is that I need a faster algorythm. I am looking at seven users of substantial complexity whose requests confound me. More CPUs will not help. I have to find a better way.

So I am following up on your comment about differences in our calculations... perhaps my algoryhtm is wasteful yet. In any case, the better solution for this problem tends along this approach:

Carve a chunk of nodes off the main chunk. Examine the number of interconnections within the new cluster, vs. interconnections back into the main chunk. Subtract the latter from the former. The result tells you how good a cut you've made.

This is apparently a hueristic approach... if I had some shortcuts to get the big picture, perhaps I could clean that sufficiently. I don't know. I'm kind of frustrated. What do you think?

[coriolinus.livejournal.com]

The long and short of this is that I need a faster algorythm. I am looking at seven users of substantial complexity whose requests confound me. More CPUs will not help. I have to find a better way.

It seems to me that what you want is a variation on the Apriori algorithm (http://fuzzy.cs.uni-magdeburg.de/~borgelt/doc/apriori/apriori.html), which is used in data mining operations to discover associations among large boolean datasets. The most common example is that it might be run on a store's sales tallies, to discover rules such as 'people who buy beer tend to buy diapers' or similar.

Another more obvious example is on amazon (http://www.amazon.com). When they tell you "People who bought this book tended to also buy these books", and present a list of other books, they used the apriori algorithm to mine that list of books.

You can find an open-source java implementation (http://users.wpi.edu/~peterg/public/random/Apriori.java) on my page, or you can go directly for the Weka Data-mining package (http://www.cs.waikato.ac.nz/ml/weka/) from which I took that file.

I know it's not a direct application; you are going to have to modify things somewhat. Even so, the algorithm you describe sounds quite similar to this, so you might be able to pick up useful optimizations by examining the code.

[chanaleh.livejournal.com]

Because I'm insane, I calculated my LJ MindMap by hand.

Proving, once more with feeling, that somewhere out there for you is a Calling that you're missing. :-)

[eleccham.livejournal.com]

And this just goes to prove the Six Degrees (or less) rule, even when the tools can't show it.

I know saxikath from callahanians, back from its Usenet incarnation. But that's not who I followed here. In fact, she would show up as Tier 2 for myself as well, were it not that I'd regrettably pulled her from my reading list for lack of time to keep up on reading LJ.

It's a very, very strange 'Net, no?

[treefox.livejournal.com]

Woah! You know chrysaphi! I love it when I stumble across friends like that in random places.

[mattlach.livejournal.com]

What do teal brackets around names mean?

[lupintonks85.livejournal.com]

Ok I made a donation so how do I know you recieved it and my mind map is one it's way in color?