A couple of weeks ago I got to talk with Steve Levine, the founder of Transparansee, a neat technology that lives on top of structured search. The model is to sell it to other sites as a custom install. Think of it as a smart layer of search on top of database-driven applications like dating, home or car buying, or, in the example Steve took me through, Fodor’s.
Transparansee’s “Discovery Search Engine” seeks to address the “stupid computer” problems which plague most structured databases. You most likely have experienced some variant of this: you put in a set of parameters meant to find just what you are looking for – for example, on Fodor’s, you want French bistros in Chelsea priced at $35 with a food rating of 20 or above – and you get no results, or only one or two. You have a sneaking suspicion that the results are missing an entire set of possibilities which are “close enough” to what you want, but you’ve been limited by the parameters you chose – if you open it up too much, you get a bunch of stuff you don’t want. What to do?
Transparensee uses “fuzzy search” algorithms to scour a database and offer on the fly weighting based on any parameter you choose. Presto, what you want to see is at hand. It’s hard to describe, but an “aha” when you see it in action. For example, there may be the perfect French bistro for you, but because it’s one block away in another section of town, it does not get found. With Transparansee, you’d see it at the top of the list, because it matches on so many of the other weights.
This is powerful stuff when you think about it, and it solves a core database search issue, at least for me: you know there is the right answer for the query you are entering, but damned if it isn’t escaping you, due to the blunt nature of structured search. Think of such a tool for Expedia, or Lexis Nexis, for example. No, I can’t point you to examples quite yet, but the site has some that you can peruse via PDF files.
It sort of reminds me of collaborative filtering, but for more types of datasets. After all, it’s hard to imagine a collaborative filtering application for home buying – “people who bought this home, also bought these homes…”.
I asked Steve what his plans were for the technology, and he said “to prove it out with as many clients as possible.” Is he open to Transparensee finding a home at one of the majors, or does he want to become the Swizerland of structured search? Too early to tell, Levine said. He’s still in early startup mode. But this looks promising, and I hope the idea spreads.
Nice coverage John.
I talked with Steve and reviewed the platform in October of last year:
http://www.buzzhit.com/2004/10/transparenseecom-better-search-through.html
Hey John,
Whats the way to send you any info on defections to Google?
I can’t find any ‘Contact Me’ link at your site. Perhaps you could put it up on your blog page.
Looks like someone finally came up with a simple solution for “no results found” — by far the most frustrating aspect of shopping for anything online. I wish there was something like that when I was searching for a digital camera online last weekend. Are there any ecommerce sites using it now? I would like to see how it worked, but could not seem to find anything like it on Fodors.
Hi I am a chinese ,and your articls are so intersted to me.Due to my poor english ,I read your articls is so hard.but I will keep on reading do my best.
Another approach to the “no results found” problem is Endeca’s guided navigation paradigm. It assumes that people are going to type basically crappy one or two word queries into a search box, so in addition to search results, you also get options to refine your query along predefined dimensions. (Also easier seen than explained).
I’ve used Endeca search and it solves the problem, but it still requires you pretty much know exactly what you’re searching for and makes the assumption there’s inventory item that matches somewhere down the path they lead you. The drill downs do give you a number of items in each category, which is helpful. But if you want to broaden your search to change your preferences to find a btter match, you have to start over again. (If you try and follow their navigation, you still may end up down a “blind alley” without finding something that matches your needs.) I like the idea of Transparansee which would let you specify your preferences at the beginning and then tells you what’s the closest match — even if it’s not an exact match to what you started with, since there are seldom exact matches anyhow.
Just looks like a multi-dimensional nearest neighbor search. Nothing new technically, but this looks like a useful set of applications for it!
Just looks like a multi-dimensional nearest neighbor search. Nothing new technically, but this looks like a useful set of applications for it!
By the way some typical pitfalls with this type of tool are prioritization. To someone on a nice date, the $$$ bistro is not at all similar to the $$ bistro next door. To someone in a wheelchair, the $ bistro and $$ bistro next door look very similar, given that proximity is more of a driving factor.
In other words, the individual users wind up having to set weights on what “nearest” means — especially as the dimensionality of the search goes up.
Where this often winds up is in an interactive parameterized search, like the one at http://www.bluenile.com
John:
I know you know that lots of people read the blog. But here’s what Steve L. told me:
“We got a whole bunch of client inquiries after Battelle