Rolling Back Search?

I’m a huge Amazon fan…most people are…what better way to avoid California sales tax on the widest range of electronics at great prices? I was interested to read today that Walmart is rolling out their new Polaris shopping search for their online properties which have always struggled to keep up with the much sexier and frankly less overweight Amazon. That’s a slightly odd name as Polaris is the name of the ICBM nuclear missile program prominent in the Cold War…the might just as well have named it their new engine “The Brezhnev” but I guess their assumption is that nobody studies history anymore so what the heck.

Just in case you care, they are featuring several things which are interesting if not necessarily useful. The first is semantic search which essentially tries to figure out what were really searching for as opposed to what you actually searched for and give you more options. This isn’t new. We were working on this back in the late Middle Ages (2002) at Fast Search and Transfer my Alma Mata of search. It’s subsequently been done a dozen different ways by just about everyone in search…but good for Walmart for catching up. The second is using social signals as indicators of quality. Essentially they collect Pinterest posts or Facebook likes along with a range of other social signals and use them as a scoring factor when delivering results. It’s not a bad idea, thou not one without potential challenges.

The premise that something is good because it has lots of likes is open to debate. Justin Beiber has over 46 million likes even Toddlers and Tiaras has three quarters of a million. The other problem with likes is that they cost so little effort to garner and are open to corruption that the moment it becomes clear how likes can be used to twist rank to commercial advantage you can be sure folks will be doing it in their droves (is that like spam?). Perhaps a better indicator is the review metric. I bought a new laptop on Amazon yesterday and the reviews were an important factor. In fact what I did was sort the result set by number of positive reviews then manually scanned down it for a laptop from a manufacturer I like at a price point I could afford. I like reviews as they are inherently more effort to carry out and potentially less open to corruption. If I went by likes I would be pushed towards Dell who has over three times as many likes as Asus. However if you compare the Amazon user reviews between the Dell equivalent of the Asus I ended buying the Asus is a hands down winner.

I took a quick swing through Polaris today and overall it’s pretty good. When I gave it very general searches like “Gold” it came back with gold jewelry, books, some music, xBox live gold membership, a blue and gold tent even a gold metal detector. Not bad for an almost impossibly vague query but in truth all over the map. What it doesn’t do is give a “more like this” option to allow you to zoom in on what you did mean. With a difficult query with relatively few good matches like “Private Parts” it returns the Howard Stern movie first then an interesting range of stuff ranging from books for children on how to avoid abusers “keeping private parts private” to Saving Private Ryan, a Shirley Temple movie and a twisted metal curtain rod where it appears to be ranking based on having the phrase “Includes all necessary parts.” In contrast the same search on Amazon also yielded a range of choices but the results presented many more choices focused on the more likely intent options those of gold jewelry and the Howard Stern Movie. That was an awfully cursory test but initial verdict? “Close but no cigar”.