Sylleptic

We all use websites that employ advanced recommendation software today. Amazon has its “people who like this also like that”, Apple’s iTunes Music Service recommends music based on past purchases, and Lovefilm uses your ratings to find people with similar tastes, to recommend movies that those other people liked.

The software that does this can be complex and expensive. We’re working with Autonomy for one client, which is very powerful and clever, and products like IBM’s Websphere Commerce Suite and ATG Commerce offer this as a standard feature for the top-end e-commerce site. Amazon devotes masses of its engineering effort to getting recommendations right. Last year, Netflix offered a million dollar prize to the person who could most improve their recommendation software.

The other day I was talking to a friend who’s setting up a property search business, about whether recommendations would be a good idea. It looked like it’d be too big and complex in the short term. Then I did a bit of digging around and came up with Criteo, which is a recommendation service, built so that anyone’s website can use it.

It looks very quick and simple to implement. You sign up for an account and then whenever a user expresses a liking for something (this could be by clicking, rating, anything you take as showing a preference) you tell Criteo. It uses a simple XML message “user xxx likes yyy”. Criteo accumulates these, and then when you want to generate recommendations, you ask it for them. It responds, again in XML, with details of all the things your user might like, based on all the preferences you’ve told it about from all your users.

Is it any good? I’m not sure yet, but it’s a very attractive idea, for anyone who wants to do recommendations on a budget. At the moment, the service is free, but I doubt if that’ll last.

In the meantime, they have a blogger’s version of the service called AutoRoll, which you can add to your blog to get a list of links to other blogs that your visitors might like. I’ve got it installed now, and I’d be interested in feedback on how good its recommendations seem to you.