I’ll be the first to admit that I’m a sucker for whiz bang tools that change the way we look at things. When it comes to search, I’m always on the lookout for a better way to scan a text listing of search results. While there have been many attempts at this (remember some of the early data viz attempts by Alta Vista?) I always ended up going back to relevance ranked text search results. Nothing can really beat the speed with which the mind can process text on a page.
Grokker.com is the latest attempt to make searching more efficient through the use of data visualization. They’ve been around for a while with a client side version of their graphical user interface and they’ve been showing off their front end to Yahoo Search since early May. After being reminded of Grokker by a post from Steve Goldstein, I thought I’d give it another whirl.
Conclusion? After 10 minutes of playing around and trying out different searches I’m not convinced that this is going to bring any extra value for me. I think that a generic search does not lend itself well to this type of presentation. The clumping of similar results into a sphere that you can explore may be helpful if you’re searching on a really broad concept (i.e. "polish") and do not know where to begin but as soon as you have a set of results focused enough to scan through them, the visual representation breaks down and is less efficient than plain text. Grokker tries to overcome this with the text preview box but the slight time delay makes it more frustrating than effective.
I am usually pretty tuned in to how I want to search on something and how to construct that search. Call me an old command line guy but I know better than to search on a single, ambiguous word and expect results that will be relevant. It’s much better for me to try out a couple of examples with liberal use of the browser’s back button than to wait for an AI program to "grok" my result set for me.
This is not to say that data visualization is all fluff – it can be very powerful when applied in specific situations. I’m a big fan of Edward Tufte and all the great work being done around the Google Maps API in combination with other data overlays (real estate listings, crime in Chicago) is clearly preferable to a text representation of the same data. Likewise, Blogpulse is trying to represent blog posts over time which is a great initial step towards representing search results over time. In each of these cases, the application of data visualization is effective because the data is already focused enough so that the graphical representation of that data is effective. In general search, there are too many questions about the base data that undermine the impact of the graphical presentation to make such an interface an effective way to navigate search results.