Tuesday, October 2, 2007

visual search engines

The visual search engine KartOO came up in a conversation I had recently, so I have been thinking a bit about the design of these kinds of tools. Obviously, as the internets get bigger and bigger, search tools are going to become even more important than they are now, both in terms of their effectiveness and their usability, so what better place for visualization to work its magic than to help people navigate this frightening morass of information? There is arguably a limit to the effectiveness of the Google-style "list of hits" method for sifting through search results, particularly if we want to understand the connectedness of concepts across the web, so the logical thing to do is begin presenting search results visually as network graphs and the like. KartOO and TouchGraph are the two most compelling attempts at this that I'm aware of on the web right now.

Briefly looking at each tool individually (searching on "information visualization"), I like KartOO's related semantic categories listed on the left, but I don't find the visual element particularly useful. Some features are confusing or counter-intuitive: What is the significance of the colored "heat map" in the background? What is it registering? It suggests that the area between nodes is meaningful, but I'm fairly certain it isn't (unless very abstractly). Also, if it is some kind of density map, its distribution should be weighted by the size of the nodes it is referring to... again, this doesn't seem to be the case. Adding to the confusion, the lines connecting nodes are only visible when you mouse-over them. Finally, KartOO's approach is to present several "maps" per search, but it isn't at all clear how those maps are related (which they presumably should be, given that they are all built off the same search terms).

On the plus side, I do like some of their use of animation, and their inclusion of symbols indicating the type of file being referenced (HMTL, pdf, word document, etc... see their help page for a list), as this can be quite useful. Unfortunately the effectiveness of the latter is undermined by the addition of page thumbnails on the icons as well, which I think does a complete disservice to the visualization. The thumbnails obscure the file type icons while simultaneously being too small to be of any value on their own.

Visually, I like TouchGraph better. The interface is much cleaner, as is the relationship between nodes. It also handles much more data on the same map, which is both a blessing and a curse. While it allows you to see much more structure, the maps can quickly get too dense to read. There are some simple filtering mechanisms provided, but some more robust options would be nice (for example, using transparency to (de)emphasize elements of the graph would be nice here, so that you could focus on certain areas while still retaining some perception of the larger whole. As it stands now, you can only make nodes visible or invisible, which can be misleading in some cases.). One very nice feature is that the logo of a website is attached to its node, so that if you know your logos, you can immediately identify various groups of pages (for instance, Blogger pages are identified by the red "B" logo, and can often be seen grouped together).

My only other criticisms are on the "search" side of things: First, "expanding" nodes only adds ten new connections at a time, but new connections don't appear to be sorted by relevancy. So, for example, the first ten connections to a given node may be largely irrelevant, but the next ten might be not be. But you won't see the second ten unless you explicitly ask for them, which you might not bother to do if the initial connected nodes are not compelling enough. This may be a technical limitation. Second, while the ability to expand nodes is nicely conducive to "exploring" your search results, connected nodes beyond the first ten don't necessarily seem to heed your original search parameters. In my case, expanding a node associated with information visualization quickly lead me to a group of pages on Indian blogging, which had nothing to do with information visualization at all. This may be by design (it does make for interesting exploration), but it seems that if I'm searching for hits related to "information visualization" I should only get results related to that term.

So, in conclusion, there are features I like and dislike about both these tools, and they are both interesting to use. But I think it's important to point out that the documentation for both is pretty weak. Neither one even documents all of their respective features. KartOO in particular fails to explain non-intuitive elements of the visual display - doesn't that defeat the purpose of a visual browser? If you want a visualization tool to be widely used, the single most important feature (besides making the tool as intuitive as possible) is good documentation. As they stand now, I get the impression that these tools are just a little too complicated or unintuitive to persuade the average user to switch from something like Google.

Am I wrong? Are there folks out there that are using tools like this instead of Google? If so, what do you like about them? What makes them more useful?

No comments: