News

Thursday, 3 September 2009

May We Recommend...

Mippin's not just about accessing the content you already know you like -- it's also about discovering the content you didn't even know you'd like. For example, without Mippin I'd never have found gems like Lifehacker (and Hollywood Backwash, I admit). If only she had the time, Mippin's own Teri would be happy to recommend excellent new sites to each and every Mippin user personally. But fortunately the machinery at Mippin comes through with tips of its own: on your My Sites page, notice that "Sites you might also like" section?

Using the same bag of tricks that Amazon uses to recommend books or DVDs based on past purchases, known in the biz as "collaborative filtering", Mippin can likewise guess at sites you will like to read, based on past visits. Sometimes they're like the sites you already read -- sometimes it gets daring and recommends a couple sites you might not normally consider reading. Try it. The more you read, the more Mippin can help, and the more quality articles will live in your Mippin homepage.

How's it done? Mippin uses software from Apache's open-source "Mahout" machine learning project (http://lucene.apache.org/mahout) to figure out what you might be interested in. I am in the fortunate position of being a part of both Mahout and Mippin, and have guided some big updates to the recommender system. Expect better and more frequent suggestions from Mippin shortly as we roll this out.

Machine learning techniques, once the territory of rocket scientists and PhDs, have become quite accessible these days through projects like Mahout. These techniques are essential to building competitive, intelligent services. Watch for recommendations to figure prominently in some next-generation projects in the works over here at Mippin.


Sean

0 Comments:

Post a Comment

Subscribe to Post Comments [Atom]

<< Home