Etsy Experiments with New Recommendation Engine

Handmade marketplace Etsy is experimenting with its own recommendation engine technology called the Etsy Taste Test. The tool, available at tastetest.etsy.com, offers you a short quiz where you click photos of things you like. You can choose to look at either photos of items for women or items for men. The end result is a list of recommended items matching your taste… in theory at least; the experiment is still very new and results can be mixed.

Still – an algorithm for artistic crafts? Sign us up!

Sponsor

How Well Does It Work?

We heard about the experiment from a post on Fred Wilson’s blog today, a VC with an investment in Etsy. However, it appears that the experiment was announced in Etsy’s forums back on Dec. 7.

So how well does it work? It’s hard to say – in tests, I was fairly happy with around 65% of the results, but then, I find it hard to not love so much of what I see on Etsy. There were some odd recommendations, though – things I would never buy for myself, not in a million years – but they didn’t dominate the results by any means.

So I like purses…

Some commenters on Wilson’s site tell a different story. Says RJ Johnston, the engine “performs as expected” and recommendations were “mediocre.” Another, Sebastian Wain, pointed out that the algorithm it uses must not be too advanced: “I chose one robot and in the end a lot of robots appeared (I just liked that robot!)” (Hmm, maybe that explains all those purses, too.) However, a third, Aynsley, says ” I finished in less than a minute and loved my results.”

The different experiences may be chalked up to how regular a customer you are on Etsy. Aynsely and I aren’t. I love to browse Etsy, but choose to spend my money elsewhere when I have it to burn (mostly on gadgets, phones and computers, to be honest). For new customers like us, a taste graph can serve as a great introduction to the millions of items on the site. Or to be more precise: Etsy says 7.7 million items are currently listed.

Etsy certainly isn’t the only site to offer recommendations these days – the Internet is using custom-built taste graphs like these (along with Facebook user profile data, of course) to build incredibly personalized, unique experiences for many Web services’ end users. Amazon has been tweaking its recommendation algorithm for years, even offering developers large cash prizes to improve upon it. Facebook recommends pages you might like, Apple recommends music and apps. My6Sense recommends what you should read. And taste graph-building technology Hunch seems to figure out your likes based on series of simple questions.  (It’s downright creepy!)

How well does Etsy do in comparison? Well, it’s a start… but there’s plenty of room to grow.

Discuss