|
This discussion has been archived.
No new comments can be posted.
|
 |
|
|
|
|
|
|
|
Wow, it's hard to believe that my project has been featured on Nooface. I have been checking this weblog almost every day, and holding my breath for a time when TouchGraph would be developed enough for me to send in a notice offering it for consideration. Well, due to the editor's excellent powers of discretion :-) I guess that have to wait no more, thanks Tony.
However, as It is, I've kind of been caught unprepared, so I think that a little further explanation of the project is in order.
I guess that for the sake of briefness I'll skip the graph layout part and get right to the good stuff. TouchGraph can be used for graph layout, but my true intent for this technology is the much more interesting field of categorization. Specifically, I am most interested in Idea categorization, though perhaps a detour to product categorization will be necessary to get the commercial juices flowing.
So what is idea categorization. Well, all over the internet we are seeing content which basically consists of a short blurb of text and some links. Weblogs are a great example of this, and discussion groups are another. So we see two models for recording the information, the journal model, and the treaded forum model. Unfortunately neither of these two structures are very good for archiving the information.
Weblogs at least have evolved to the point where categories, and, even better, multiple categories can be attached to each post. But not discussion groups. The only bit of meta information that a post to a discussion group has (besides the author + date) is a link to the parent thread. Once the discussion in a given thread has died, the thread is rarely heard from again, meaning that the content therein is basically dead.
Sure, search engines can solve some of the problems with resurrecting outdated information, but they are only effective when searching for a specific keyword, or combination of keywords. What if one wants to search for a more general concept such as "open source is good" Chances are that a search on the topic will reveal too many sources (try it on slashdot). The thing is that is easy to construct a very valuable, but keyword free argument for or against the goodness of open source, an argument that is doomed to be ignored + repeated many times for the simple reason that it can not be found.
So... the solution to all these problems? Well, faceted classification can go a long way towards solving the problem of weblog archiving. The trouble with faceted classification however, is that all the facets must be explicitly named. So, a new facet must be created every time a new relationship is found between two items. And, even if it turns out that the facet is never again used to classify any other item, from it's creation onward it will take up valuable screen realestate by being offered up as a choice.
So the solution? Anonymous facets or links. Items can be linked to other items based on similarity. Later, if it turns out that a significant number of items share the same property, this property can be named. (There is also the possibility of having hidden facets, but these would just function as named links.)
Basically, unnamed associative links are very useful, especially for the case of discussion forums where people usually just want to type a quick response. People don't even like to put descriptive subject names on their posts, so asking them to name their links is a bit much.
The conversion from associative links to facets is where graph visualization comes in. By visually depicting the link structure it becomes possible to see clusters of items that would not have been recognized from a representation which only shows a node and it's immediate neighbors. An associative structure provides a fertile ground for a second pass over the information where facets could be formed.
Take Slashdot again. Hundreds of people read the whole discussion and yet no one is able to modify the tread structure. A graph-based rather then thread-based discussion would allow for readers to make a simpler contribution then adding a post, by adding a link between two nodes. Perhaps facets for the links could also be proposed and voted on. Eventually, instead of an unstructured flow of information, the result would be a nicely categorized set of ideas. The best part would be that older posts could be brought back into the current discussion by linking to them.
So how does TouchGraph fit into all this? Currently the technology is still in it's very early stages of development. Still, the current version 1.20 of the TGLinkBrowser ( http://www.touchgraph.com/newLB/TGLinkBrowser.html ) can already be used to render a weblog archive as an associative graph. Initially, all one sees is a graph with the post subject as node labels. Pausing the mouse over a node causes a window to come up containing the html of the posts. The advantage of the popup technique for browsing a weblog, is that there is no flipping back and forth between the subjects and the post bodies. To get an idea of what a rendering of such an archive would look like, you can take a look at a screenshot of the Langreiter.com Wiki ( http://www.touchgraph.com/Langreiter_TG.jpg ) Langreiter.com used a pre html popup version of the code, but you can see how the popups would mostly eliminate the need to go outside the graph.
The LinkBrowser also has a feature enabling one to link to graphs on other servers. For now one can only link to a whole graph, and following such a link replaces the current graph on screen. In the planning stages is the ability to link to individual nodes in other graphs. Following such links will result in the attached node being shown adjacent to the current graph, making it appear as if all of the remotely stored components are actually part of one large graph. Possibly such a set up could eventually lead to a massively parallel effort to categorize all of the associated information, with decentralized control by the hosts of the individual sites. This might sound like science fiction, until one considers that something pretty similar is already going on in the Weblogging community. The only difference is that the links are never made visual, and the associations in the text are never elevated to the status of facets.
Speaking of the Weblogging community, another TouchGraph application that might be of interest, is a map of weblogs created in conjunction with http://www.metastatic.org You can see it here: http://www.touchgraph.com/TGLinkBrowser_PoW.html I had to abandon this project before polishing it off completely to work on the TGLinkBrowser. Still, the results are pretty interesting. Not all the links between weblogs are shown, only the first few from any given site. Right click on a node and select expand to see all of them.
... Whew, well, that concludes this rather lengthy post. I hope that it's a stimulating read for the visitors of this excellent site. For more information stay tuned to the TG News, http://www.touchgraph.com/news.html (sorry, there is no mailing list yet). Also, if you liked the ideas expressed above then the following thread might be an interesting read http://sourceforge.net/forum/forum.php?thread_id=1 17442&forum_id=95693
Tony, thanks again for the support + exposure,
--Alex Shapiro
alex@touchgraph.com
|
|
|
|
|
|
| |
|
|
|
|
|
|
|
|
|
Distributed categorization, Exactly!
So you don't think that people would like to categorize posts?
My reason for thinking that they would, is based on the way that I read slashdot. Once you (I) read a few threads, you start noticing that people repeat themselves. Often, you are not sure, and then you actually go back and re-read the previous post. So, I think that if the reader had the ability to create a link back to a previous post, they would.
Here is another way to look at it: The act of noticing similarites (making mental associations) between posts, and the act of evaluating a post as good/bad are both reflexive actions. If you are an intelligent human being then you automatically perform these two actions when presented with a piece of data.
But why take the time to act on the impulse and actually rate a post/make a link. Well, the aswer for rating is more ovious, since one can actually experience such a system. The thing about ratings, is that they don't just let others know your opinion, but they actually effect the structure of discussion. Users can filter posts to see only the highly rated ones. Even if they don't technologically filter them, the filtering is done internally by paing more attention to higher rated posts.
The reward for linking is that with a graph-based interface, creating a link will actually change the structure of the discussion. Adding new links will actually change the flow of attention of the readers to come. Honestly, I myself an not exactly sure of all the ways that links can be used in a discussion (blogs are more obvious), but I feel that the experiment has to be made. The thing is, that this stuff is so hard to imagine that you have to build it to actually see the results.
There are times when a link could be just as effective, if not more so then a rating. For instance, suppose that someone presents a point that has already been brought up and nicely contradicted in an earlier thread. One could just rate the latest post down, but a better solution would be to add a link to the contradiction. Somehow it seems that doing so by adding a graphical link would be easier then taking the time to reply and say "hey look back there" (and provide a link).
------
Everything2 is a great idea, but I think that the site is a bit link happy. The thing is (yes, yes, another thing) is that linking blind, i.e. in a non-graphical environment, means that if there are five memes all related to each other, then chances are that each will have links to the other four. With a graphical approach, this "clique" would be rocognized, and could be resolved by a central category node linking to all five, thus reducing the clutter at each node.
------
Faceted classification, by the way is the name of the next great thing that is going to hit the internet. This method of organization is a level of magnitude more powerfull then the current mostly hierarchical scheme that yahoo's got going.
A great starting point is Peter Merholz's site, which is where I first heard of faceted classification. Check out these posts:
an early thread ,
my responce to that thread ,
a later thread on flamenco (hey hey, I mentioned it first!)
Ok, great, another long post. Well... I enjoyed writing it. Thanks for your interest, Greg
--Alex
|
|
|
|
|
|
|
|
|
|
|
|
|
Alex, I'm very excited to see the work that you're doing. One of my main interests in grap-based interfaces is how to keep the "clutter" down of all these linked facets by means of abstraction.
I belive that unlimited linking of similarity between information media has an incredible future for education (where students would be able to relate their insights to each others) and business (facets could link between possible partnerships and channels). I see an open standard for graph interfaces as being a prerequisite for these fruits coming to light.
One of the big challenges will be designing an interface that allows for easy linking between these information nodes-- being able to make selections that are to be faceted in their native media format, designating them as a facet, and them connecting them to another information node. Since any selection of content could become a node, there will be incredible glut of information inbetween making a simple link of similarity.
Beyond the scope of what you've written here, I'd like to know what you think about information processing and problem solving once there is a significant pool of data.
In the past I cofounded and worked at a company that tried to do all this for the graphic design market-- it ultimately did not succeed, but I wish you better luck.
Cheers,
Matt Weber
Weber Studios
|
|
|
|
|
|
