• Home
• About
• Writers
• Landscape
• Wizards Index
• Fraud
• Wanna Be Like Larry?

Recommendation Engines May Engineer the Soul

“Recommendation engines aren’t designed to give us what we want. They’re designed to give us what they think we want, “ says Lev Grossman in his recent Time Magazine article, If You Liked This… . And that’s the crux of the difference between recommendation engines and perfection.

In my perfect world, I would open a retail store called “YOU” and you would shop there all the time because every product in the store would suit your taste. I would use your buying habits to build my inventory. You would spend almost all of your money in my store. You would be happy and I would be rich. Fair trade.

In a sense, that’s what recommendation engines do. They use what you’ve already purchased to guess what you might like to buy next… and they offer it to you immediately. It’s you recommending something to yourself with the computer as an intermediary.

Word of mouth from a friend is, by far, the most relied upon source of confidence, says a recent survey.  Statistically, almost 90% of us trust a friend’s recommendation to some degree. So wouldn’t you assume that, by being your own best friend, you couldn’t go wrong? You know the answer to that loaded question.

“The trouble with recommendation engines,” says the author, “is that they’re really hard to build. They look simple on the outside—if you liked X, you’ll love Y—but they’re actually doing something fiendishly complex. They’re processing astounding quantities of data and doing so with seriously high-level math. That’s because they’re attempting to second-guess a mysterious, perverse and profoundly human form of behavior: the personal response to a work of art. They’re trying to reverse-engine the soul.”

Technology can engineer one’s soul directly or indirectly, the addled goose assumes. Source: http://www.gizmowatch.com/images/bci_48.jpg

A lot of companies are trying hard to link one preference to others but, unlike the alphabet, human beings just don’t go from A to B to infinity and beyond in any algorithmically defined order.

Pandora, Netflix, Amazon, Facebook, eHarmony, MySpace and the like are tying hard to get it right. Industry studies show that about a third of us buyers choose another selection from the recommendations, so the value is obvious to both merchant and buyer. But getting it right is proving much more problematic than anyone thought.

It can be very labor-intensive. Pandora music experts analyze 10,000 new songs a month– a process that results in a numeric rating for the song’s many attributes–as a way of tying one selection to others like it. It then uses this data to link your song choice to the one you may want next… and offers you the chance to buy that next song (or songs) while you are still on-line. It works, doesn’t it? The Pandora library has profiled about ¾ million songs to date.

Netflix has, for three years running, offered $1million to the individual or team that came up with a better recommendation engine than its existing model, a prize recently awarded to BellKor’s Pragmatic Chaos team for beating the Netflix standard by 10 percent. Imagine… a guess that is just 10 percent more accurate is big bucks all around. One Forrester Research industry analyst says recommendation systems can account for 10 to 30 percent of an online retailer’s sales.

Just one generation ago (in 1994), says Sheena Iyengar in her book, The Art of Choosing, there were only about 500,000 different consumer goods in the U.S. marketplace. Today, Amazon alone has more than 24 million products to sell. Have we become buying machines or what?

Choosing is tough. Help is welcome. Recommendation engines come to the rescue, but they are not perfect.

One weaknesses, says Grossman, is their inability to recognize that, just because we liked Rocky IV, we might not be that anxious to see other movies featuring Dolph Lundgren… or that our next book after The Girl With the Dragon Tattoo is not, logically, Pippi Longstocking. Logic does not always prevail when it comes to putting two purchases together. The human propensity to change our minds, want something different or think out of the box drives computers wild. If you buy three jazz tunes in a row, it thinks you are Dizzy Gillespie or Jelly Roll Morton and recommends accordingly.

Not unlike the quest for an artificial intelligence machine that acts like our brain, recommendation engines haven’t got us figured yet… but they are getting better.

Where there is a market, there are suppliers to serve it. Can’t put your own recommendation engine together? There are companies that would love to help. ChoiceStream (the largest with more than $10 million in revenue), Aggregate Knowledge and Clever Set have different methods for getting to a similar end, and they are ready to customize your base and improve your revenue per sale big time.

Despite their inhuman weaknesses, what I like best about recommendation engines is that they make buying more fun. Tempt me! I love it. Recommendation engines almost always deal with what appeals to out pleasurable wants, unlike search engines that work to serve our intellectual needs.

While recommendation engines still have a way to go, the intensified quest for the next Google, Bing, Yahoo, etc is quite another thing… infinitely more complex because in your need for an answer, they draw from the universe instead of just little old you.

Jerry Constantino, Contributing Editor to Beyond Search, July 28, 2010

Comments

Copyright © 2008 Beyond Search · ArnoldIT.com · Disclaimer · Revolution News theme by Brian Gardner · Log in