Ask: Complicated So It Must Be Good!
Published: May 25, 2007
Author: David Rodnitzky
Ask.com is spending millions on TV, billboards and newspaper ads proclaiming “the algorithm killed Jeeves” among other things. No doubt, the message of these ads is that Ask uses very complicated technology to provide great search results.
This message was definitely drilled home with Ask’s full page ad on page B5 of today’s Wall Street Journal. While I couldn’t find an image file of the ad online, the ad headline pretty much sums up the concept: “The Algorithm Sees the Internet The Way Dmitry Sklyarov Sees a Poorly Encrypted DRM File.” The ad also includes several mathematical formulas, confusing charts, and seemingly random inclusions of words like “vacuum energy density”, “Riemann Hypothesis”, “autosomal recessive gene” and “bibliometric.”
But there is also a story in all the multi syllabic text, and the story basically goes like this: the major search engines are dumb and Ask is smart.
Assuming any WSJ reader other than myself actually read all 500 words of text, I could see the high-falooting wordsmithing working its magic. You might end up thinking “gee, I don’t understand any of this, but it seems that there is a lot of neat technology behind Ask. I should check it out.”
Unfortunately for Ask, I actually did understand their argument, and I came away with two negative thoughts: 1) The ad makes many false accusations against Google and Yahoo and 2) It sounds like Ask hasn’t improved their algorithm much since their acquisition of Teoma.com may years ago.
First, let’s discuss the claims about, as Ask calls them, “today’s most widely used search engines.” Here are a few claims that are flat out wrong:
1. Claim: “They find sites with the most links and present them as authorities. This is roughly analogous to handing the Fields Medal to your high school homecoming queen.” The Truth: Google counts the number of relevant links to a site as part of its ranking algorithm. So, for example, if you had a Web site about homecoming queens, inbound links from other homecoming queen Web sites would be relevant, but inbound links from sites about the Field Medal would not.
2. Claim: “Counting inbound links isn’t enough . . . the heavy hitters of search all use the same mathematically myopic approach – counting links back to authoritative Web pages.” The Truth: Well, for one, this directly contradicts the first claim above, that Google uses a pure “link popularity contest” to rank pages. But in addition to that, the claim above suggests that Google adds the same weight to each link. As suggested in the original Page Rank algorithm, this isn’t true.
3. Claim: “And never again will you get “results” consisting merely of ten blue links, rather than the rich aggregate of images, video, conceptually related search topics and pure expert insight the Algorithm delivers.” The Truth: Three words: Google Universal Search.
OK, so Ask is lying about their competitors’ algorithms. I can dig that – this is an ad after all. But what does it say about this amazing new algorithm. Let’s deconstruct that as well.
1. Claim: “It works like this. For each search query, an index G of Web pages is found. For each page p, you associate a non-negative authority weight a(p) and a non-negative hub-weight h(p). This will lead you to the rather obvious conclusion that when p points to lots of pages with big a values, it should get a big h value (inverse weighted popularity). And when p is pointed to by lots of pages with big h values, it should get a big a value (weighted popularity). ” Explanation: Ask looks at a page in the context of a community of other relevant pages, and rewards higher value to pages that are more respected within a particular community. This is the entire concept behind Teoma, and as far as I can tell, there is no difference between this explanation and Teoma of, say, three years ago. So the Algorithm was developed several years ago. Woo-hoo.
2. Claim: “But while you are learning from the Algorithm, the Algorithm is learning too. It studies the way anonymous groups of users search and forms an aggregate view of which results those users find the most valuable.” Explanation: Well, this is basically the same as the point they made above. This ‘learning’ algorithm is nothing more than a collaborative filter – the same technology used to power Amazon’s “people who bought this also bought,” Yahoo’s launchcast, and Web 2.0 companies like delicio.us and Flickr.
3. Claim: “It’s here to narrow or expand your search based on concept – something no other search engine can do.” Explanation: I actually had no idea what they meant by this, so I decided to do a search on Ask for the keyword “love.” On the right navigation, they have a list of related terms, some very narrow, like “love poems” and some quite broad, like “Hearts,” “Hate” and even “Winnie the Pooh.” I did the same search on Yahoo and got an “also try” for “love quotes”, “love poem” and “love calculator.” On Google, on got nothing other than search results. So my conclusion on this point is that Ask has done a better job of integrating semantic clustering into their search results, but this isn’t going to revolution search relevancy.
I have no doubt that the good folks on Madison Avenue who created this ad never expected search nerds like myself to deconstruct their work and expose their facade of intelligence. But, as I often remind myself, the number one search on Google is “Yahoo” and the number one search on Yahoo is “Google,” which goes to show that the average American doesn’t need much convincing to be convinced that search algorithms are complicated.
Don’t worry though, Ask, I can assure you that the average American doesn’t read this blog, or any blog, or even the WSJ for that matter. Your secret is safe.