Suggestion Software: Perfect match for idle CPU cycles.

I recently tried a silicon oracle called GetGlue.  I’m genuinely impressed at how racks of CPUs can quickly navigate through an enormous knowledge graph and grab a suggestion node that matches some characteristic of my preferences. Great searching capabilities, but I’m less than excited about the results.

Like other recommendation software (say, Hunch), you register preferences by initially rating a sample list of movies, books, TV shows, and music. I told GetGlue that I liked The Breakfast Club, Goodbye Columbus, The Great Gatsby, and Nina Simone.  GetGlue quickly responded with a list of predicable tips: lots of Philip Roth novels, Catcher in the Rye, American Beauty, and Sara Vaughn.

I suppose if I had lived in vault, then some of these suggestions would be novel. One quibble for the GetGlue crew: how about adding a “like, but already know about” classification.

On the plus side, GetGlue deserves credit for bringing swing band leader Jimmie Lunceford, who I hadn’t heard known about, to my attention.  Thanks.

Beyond the human-to-algorithm communication, GetGlue does provide more traditional person-to-person interactions, which may be the most valuable aspect of this service.

As Mr. Shirky has pointed out, social networking software taps into our cognitive surplus, wasted brain waves that were formerly devoted to watching TV, and harnesses that cognitive energy for more useful activities. Need travel advice? TripAdvisor.  Restaurant review and food photos? Foodspotter.  Fashion advice? GoTryItOn. General expertise? Aardvark. What is everyone doing right now? Hot Potato.  And on and on.

What I’ve discovered is that suggestions received from this altruistic crowd—special kudos to TripAdvisor and MetaFilter—are far better than tips dispensed by a suggestion machine.

I was sucked into the Hunch vortex a few months ago, and tried out this heavily publicized recommendation seer.  I engaged in  a t e d i o u s training process, which was not worth the trouble judging by the results.

You might as well  devote the time spent on training a slab of silicon by going to book group or taking a class or visiting a museum or reading a special interest magazine or, ahem, a blog.

I just don’t believe that CPUs crunching through algorithms based on probabilistic distributions  or contrived distance metrics (read k-closest) will be able to give much insight into wines or poetry or works of art.

Recommendation analysis may be a good use for wasted CPU cycles—kind of like harnessing idle PCs to calculate pi to the millionth place—but the results I think are mostly a curiosity at the current state of the art.

Thankfully, GetGlue gives you the opportunity to see the likes and dislikes of other people in your social sphere.  There’s nothing unique about list making on the Web, but it’s nice to see that  GetGlue has hedged their bets on mechanical oracles with the advice and experience of humans.

Enhanced by Zemanta