I'm not sure how to resolve this problem...
TouchGraph works best when it's displaying fairly simple data structures. Once nodes have more than five edges(links), the graph becomes muddled.
memes.net works best with a high degree of intertwingularity. The more relationships between nodes, the better.
What are your thoughts on intertwingularity vs clarity? Perhaps the discipline imposed by reducing the number of links can useful in itself. Is there any way to bridge the gap: make a highly intertwingular repository also have a high degree of clarity when presented visually?
(Here's a bit of a starting point if you've never heard of intertwingularity: http://www.memes.net/index.php3?request=displaypage&NodeID=323 . The term was coined by Ted Nelson I believe.)
Perhaps if a node has many edges, and is linked to another node with many edges, the tension of that particular edge should be relaxed a bit.
This might spread out the large clusters, while still expressing the relatedness aspect.
I should be able to tweak this in my php code for www.memes.net
I did this. I reduced the tension whenever two nodes with many links connect to each other.
I'm not sure if it helped though...
It makes small repositories look worse, and large repositories look better.
http://www.memes.net/index.php3?request=exportTouchGraph looks a little cleaner
http://ms.memes.net/index.php3?request=exportTouchGraph looks a little sloppier
Interwingularity. Hmmm. In-ter-wing-u-la-ri-ty
My thoughts on Interwingularity vs. Clarity is that clarity is better, and that for mind-maps at least, the discipline of keeping the number of liks down is very benicial.
If you can't break a group of more then 9 items into sub-groups, then chances are that you yourself are confused. (As I found was true for me)
Personally, when I used TheBrain as a mind-mapper I had a rule about trying to keep the number of links down to about 5. I think that it's a good rule for mind maps, because of the magic number 7 +/- 2 rule. See http://www.well.com/user/smalin/miller.html
The idea is that a person can't keep more then 7 items in mind at a time. The thing I like about networks, is how easy their structure is to remember. I am afraid to loose that property if the degree of branching exceeds 7.
Also, my attention span is often not long enough to sort through long lists (when I'm browsing for pleasure). If I don't find something I am interested in within the first seven items, I'd rather go back where I came from. (Talking from semi-experience).
As for representing thigs visually, I think that you are being harsh when you say that nodes with more then 5 edges become muddled. I think that number is more like 10. There is a bit of an overlap problem, but I think I can fix that. If every node had 10 edges things would get hairy. If only a few do, and there are lots of leaf nodes attached by a single edge, then things should look ok.
Also, the reason that your examples look confusing, is not a high degree of interwingularity, but because there are simply too many nodes. I am pretty that things will look a lot better when locality is imlemented.
By the way, now that I think about it, did you fix the www.memes.net graph? It really does look much cleaner then the last time I saw it. You reduced the number of nodes, right?
I still have the same number of nodes on www.memes.net, but I greatly reduced the linking between nodes.
I may need to come up with a method to determine which links from Lucid Fried Eggs should be represented in TouchGraph. I could ascribe a relevance factor to the links, or just let users decide on the tension themselves when they create links.
Consider other html repositories. Pages like Yahoo become less useful if only 7 subtopics are available per topic. This is partly due to the nature of html: click/wait/click/wait. It's best to have many links, resulting in fewer clicks.
As an exercise, it might be interesting to try and display portions of the open directory project, www.dmoz.org in TouchGraph.
Yeah, I guess you could let users decide for themselves. My initial instinct is to trust the average user, though I have no experience with running discussion groups. Why not have users rate the relatedness of their link on a scale from 1 to 5.
As far as the click/wait on yahoo is concerned, this is a good point. If you have to click and wait. But, the advantage of a graph, is that you have a much broader view of things, so you have to do a lot less clicking and waiting.
If all nodes have a degree of 7, then you can easily display all nodes that are 2 steps away, 49 of them. If you have a degree of 5, then you can display nodes that are 3 steps away (125).
Actually, rereading your post, I think that you knew this already. So you are right, the reason that things are the way they are now, is because having only 7 links per category would produce a lot of clicking and waiting. Another reason, is that the dmoz is mostly a tree structure. There is no multiple inheritance. And if there is no multiple inheritance, then you have to put your items in small categories. Or be subjective as to the smaller categories that you chose (since the possibility of multiple inheritance indicates that there is more then one way to break the nodes into small categories).
We could display the dmoz, but I don't know if I want to. I would rather leave games like that to TheBrain. The open directory project does have too many nodes per category, and that messes with the way TouchGarph displays things. I would be more then interested, however, if someone besides me wanted to do this experiment. Just to have a look. But personally I don't think I would use the results.
Thinking about it, what I really do want to do it to graph the goodle "find similar website" search results. The really cool thing about those is the lack of simmetry, the #1 similarity for one item is not necessarily it's #1 link. I bet that goodle stores the degree of similarity too. We could only draw edjes to the first fiew items, and make the other edges invisible. Ahh the possibilities.
And if there is no multiple inheritance, then you have to put your items in LARGE categories
You might find this approach interesting.
This project is built on top of TouchGraph.
When I display node, I use this logic.
1 node that is the selected node.
7 nodes are linked to this node. (The 7 nodes that are the "more similar" in my case.)
3 (*7) new nodes are added to these 7 nodes
ect... This addup in clarity.
You can have different view.
Log in to post a comment.