I’m not sure how I feel about this – maybe the magic is good magic, maybe it’s voodoo magic, or maybe it’s fake magic, the work of a charlatan, but I wonder, I wonder, might Google’s ‘Personalised Ranking’ utility in Google Reader be useful in filtering, or at least ranking, latest issue table of contents feeds from somewhere like TicTocs?
Only have a 10-minute coffee break and want to see the best items first? All feeds now have a new sort option called “magic” that re-orders items in the feed based on your personal usage, and overall activity in Reader, instead of default chronological order. Click “Sort by magic” under the Folder Settings menu of your feed to switch to personalized ranking. Unlike the old “auto” ranking, this new ranking is personalized for you, and gets better with time as we learn what you like best — the more you “like” and “share” stuff, the better your magic sort will be. Give it a try on a high-volume feed folder or All items and see for yourself!
[Google Reader Personalised Ranking]
Now I believe that there is also a JISCRI project looking at a related sort of thing – Bayesian Feed Filter…: “The Bayesian Feed Filtering project will be trying to identify those articles that are of interest to specific researchers from a set of RSS feeds of Journal Tables of Content by applying the same approach that is used to filter out junk emails.” [Project Kicks Off]
So I’m thinking: it’d be great to see how their approach might filter subscribed to feeds bayesed (!;-) on what users read from those feeds, compared to the Google magic?
Thanks for this post, I hadn’t come across Google Magic. It does sound a lot like our Bayesian Feed Filter, and I guess it works on a similar type of magic. We’ll be releasing info on how well our version has worked soon, we’ve just got one or two or our trial users to chase up and some data to crunch. It’s not easy to see how we could do a direct comparison with Google Magic on that data set (it’s RSS, ephemeral, so the items that were used for training during our trials have probably gone) but we’ll give it some thought. Anyone interested in a direct comparison from scratch is welcome to create an account on our Feed Filter. Details in the post at http://bayesianfeedfilter.wordpress.com/2009/08/25/about-sux0r/ (which also has some comments about why the Google Magic approach might not be well-defined enough to work … unless Google are really really clever with lots of data at their disposal)
I wonder if this sort of bayesian filtering would work for other datasets?
For instance, in a digital library. A student (or anyone else for that matter) could choose the types of books or resources (videos, etc.) that they enjoy reading or find useful.
After a bit of training, the system could recommend new, relevant material that they might enjoy.
I think that such a system could have the potential to give exposure to otherwise undiscovered learning resources, while at the same time increasing the quality and quantity of the reading performed.