Making Search Better (Part I)

Like you, I've been an avid search engine critic for years. When I don't see results I like - be it on Google or someone's ecommerce site - I squawk and complain like the next person. Sometimes I even fill out the feedback form (did you know you can click on "Dissatisfied? Help Us Improve!" after you do a Google Search, and tell them why you didn't like the results?).

In rare cases, I blog about how dissatisfied I am, or write an article about it. Obviously I'm more passionate about searching than most people!

So imagine my feelings when I had the opportunity to work on improving the search engine for the HomeStars site. Not as thrilled as I should have been! I thought it would be like getting handed the keys to a Ferrari. But then it dawned on me: I've been invited to the Ford plant and asked to improve the way they manufacture the pistons.

The history of information retrieval runs deep, but the history of web information retrieval has been short, and it has to be said, kind of funky. In either case, some of the people who designed very popular forms of search -- think Yahoo directory -- had no business doing so. If you're anything like me, you've had bouts of extreme satisfaction with information retrieval (I love Google Maps to save my bacon), as well as some years of extreme tedium. I sometimes think that my one main reason for specializing in democratic theory in graduate school was so I could just stick to the same shelf - JC 423 - in the stacks. I'd wander up and down those aisles, looking for any books I hadn't yet read. In the process, I discovered some pretty mind-blowing stuff. :) It makes me wonder if future generations will have a similar experience. I have to admit they probably won't, and it won't be the end of the world.

As imperfect as search is, it always improves.

In the case of groundbreaking technologies like Google, they've often been built on one big idea, and stabilized around the real-life improvements that come with tens of thousands of small refinements.

Search engines do their best refining their proprietary sorting recipes with the help of user data in various forms. Do people find what they need on the query "joe's plumbing" or is something messed up in the results? Feedback helps us improve.

Obviously when Google sifts through feedback, they are largely looking for the high-level stuff, like people being frustrated by not finding what they need. Search engines are businesses, not purists. While companies like Google are lucky enough to have a campus packed with people who can spend 20% of their time on pure discovery, the main thing they do would seem to be to obsessively check up on whether aggregate satisfaction scores are improving.

You cannot satisfy every single user for a couple of reasons.

The first reason is the "universal search box" concept. Search engines these days have trained users to just "go type in the box." That means no matter what kind of search, or no matter what intent behind the search, the sorting algorithm has to pull out the best possible information for that search, without really knowing who you are or what you want. So in other words, a fifteen-year-old typing in a query about an anime cartoon (Bleach) is treated the same way as a 43-year-old parent of four who is looking for info about Clorox. Search engines are only as good as what you put in!

And search engines are not mind readers (though they're improving on that score, through personalization). They're machines. The best ones can try to guess a little bit better from what you type, or where you're located. Failing that, they should at least work in a relatively predictable fashion.

Second, related to the "go type in the box, which is supposed to solve every type of problem" phenomenon, is the fact that people don't want to become better searchers. Search on Google has gotten so good, no one really expects to have to "learn to search better" anymore. And that's a shame.

Twenty years ago, the only folks doing any "searching" were pretty much undergraduates checking out the university catalogs, and the odd soul in the public library. The mainstream public started trickling through the doors in the early 1990's about when Webcrawler and Magellan were available on a text-based browser, and they came along in larger numbers after the growth of search engines like Yahoo and AltaVista.

Today, everyone knows how to "search"... sort of. Search has gotten so kickass good (ever since Google's cocky "I feel lucky" button) that few users use advanced search options.

So I guess we can forget the idea that the average person will bother to read the "instructions" on how to search better. So we, the search interface designers, just have to suck it up and make it work - as challenging as that can be.

This "Google effect" is similar to the bar-raising "Amazon effect" in online retail. In retail, if your shipping is expensive or checkout appreciably more difficult than Amazon's, you're dead in the water. It means all of the little guys have to work harder.

Luckily, we already know a fair bit about what people will be searching for within our database. It's different from searching the whole web.

In fact, compared with smaller more focused (sometimes called "vertical") search databases, the concept of searching the entire web is ... well... silly. The fact that Google has to go to such elaborate (and often inaccurate) lengths to determine which pages have "authority" and "relevance" is proof that the haystack they're helping you find that needle in is just too big. Billions of pages!

With only a few hundred thousand, or a few million, potential pages, the job gets easier. Vertical sites (like HomeStars) have their own industry-specific ways of ranking and rating stuff that is much more reliable because the community there is more dialed in on what counts.

Well, that's a heckuva preamble. I'll share a bit about what we're specifically trying to do to improve the search experience at HomeStars, in my next post on the subject.

Labels: search

Posted by Andrew Goodman
2 comments | Permanent Link