A9.com, Amazon's web search startup, has done a major design that completely changes the site.
It now appears to be a metasearch engine, like Dogpile or Metacrawler, that gives a lot of control over which seach engines are used.
I actually like the site better than before -- it seems cleaner and more usable to me -- but the functionality seems minimal, merely showing search results side-by-side from many search engines. Despite the spiffy AJAX UI on top, this is the kind of thing that has been around for a decade.
I would love to see A9 go a big step further and automatically decide which of thousands of search engines to query based on the information need of the searcher, then combining and reranking the results. That is a hard problem, but a very interesting one.
Danny Sullivan has some scathing comments about A9's redesign. Danny says, "Frankly, A9's always felt like some type of Amazon plaything, a way for Amazon to say they were in search but also pretend it was all just an experiment, if it failed to succeed. I think the failure is now apparent, and Amazon seems to be cutting its losses pretty dramatically."
Ouch, but there is a lot of truth in Danny's words. A9 spent millions going nowhere instead of attacking the interesting and hard problems in personalized search, federated search, and personalized advertising.
See also my previous post, "What will become of A9?"
Update: The punches keep coming. Paul Kedrosky says Amazon has been "wasting their time in nowhere search efforts." Joe at TechDirt says, "[Amazon] realizes it doesn't have much to bring to search." Ouch.
Update: Six months later, it appears this redesign did nothing to prevent a sharp drop in traffic to A9.com.
Monday, October 02, 2006
Subscribe to:
Post Comments (Atom)
5 comments:
Since we are at the topic of search engines, let me throw out my views and see do they really make sense.
I want to see the future generation of search engines to support clustering of results rather than personalized. I am not sure if it really make sense to personalize the search results. Say I am searching for "Data Mining", can any one personalize the results? If yes, what kind of personalization would it be? Is it something like: Since I searched and clicked on "classification" related links (in history), do you expect the search engine to give me "classification" related links for the query "data mining"? Then, user is left with "classification" results and nothing else. It doesn't really make sense, does it? The ultimate goal of search is to give whatever is appropriate to the given query.
Instead, if the engine can give clustered results (something like grokker did), then user can effectively choose from the different clusters he see.
Even if personalization make sense,
its applicability is limited and
I think techniques like clustering should be given more importance compared to personalization. Furthermore, if we had the clustered results at hand, then one can do better job in personalization. Don't you think so?
What say u?
Hi, Shirish. It is a good point and, given that Raul Valdes-Perez agrees with you, you are in good company.
I have some thoughts on your and Raul's argument in an old Nov 2004 post, "Personalized search vs. clustering".
Please let me know what you think.
I thought the a9 search engine homepage was too complicated to be popular, and as other seach engines one by one fall in line with a simplistic start page a la google, we see that it's a deterent to people, so the redesign was necessary. Still, the options it offers won't fit, I think, with common search users, and I notice one big feature that got the cut, a9 maps. a9 maps had a really cool "drive through the street" feature that I'd been using of late to view an appartment before i rented, and look up stores on my credit card bill. No other engine I know of offers this feature, with the acception of microsoft, that has that test two city drive through page, but I can't even find it easily going through live.com. Does a9's deal with microsoft have anything to do with the cutting of a9 maps? (i assume they have a deal since live.com is the cornerstone of their metasearch engine)
I agree with shirish. And the more I read the things Greg has to say, the more I agree with him, too.
As Greg and I discussed once before, why not a compromise between personalization and clustering? Why not have "personalized clusters"?
The thing that Vivisimo does right is that they don't get in the way of the regular results. Since the clusters are off to the side, they are no more distracting than Google ads. They're there if you need them, and not there if you don't need them.
So why not have the initial search return the personalized results in the main column...and then have clusters on the left, for further modification, if the personalization got the current need wrong?
Even better, since there are multiple orthogonal dimensions along which you can cluster data, why not make the clusters themselves personalized? Look at some of the main dimensions along which a user tends to browse/search, and cluster the -current- query along those historical/personalized dimensions.
You have these companies like Google that are so anti-clustering on one hand, and others like Vivisimo that are so anti-personalization. I see room for compromise.
Yep, it likely will be some combination of the two. I agree, Jeremy.
I love Clusty and Vivisimo, by the way, and have said so many times on this weblog. Fun and useful tools.
Post a Comment