Comments on: Who will filter the stream first? http://www.craigekerstiens.com/who-will-filter-the-stream-first/ My thoughts and predictions on technology and business, and sometimes strong ones at that Tue, 26 Jan 2010 23:01:25 -0800 hourly 1 http://wordpress.org/?v=3.0.1 By: Craig http://www.craigekerstiens.com/who-will-filter-the-stream-first/comment-page-1/#comment-25532 Craig Tue, 26 Jan 2010 23:01:25 +0000 http://www.craigekerstiens.com/?p=160#comment-25532 I agree it COULD result in mis-classification, but there's a lot being done quite well with unsupervised clustering that's proving effective. Users on at least some services such as twitter are already providing tagging of other users, and validating which items fall in a category via tagging. By expanding the classification to just beyond California, to California and Traveling you could clearly identify that as an area of interest, thereby eliminating things only pertaining to California and only pertaining to Traveling. Essentially any filtering you could do would be some improvement, and how effective you could be would be more restricted by the amount you wanted to see in your feed as much as it would on how effective your filtering was. Only for the users that greatly limited what they wanted in a feed would you miss out. Since you're already opting in to relevant users, instead of on the entire corpus of twitter you have to worry less about the overload of information of what may occur for conflicting classifications. I think your initial question really becomes the driver for then being the input to filtering the stream. You begin with unsupervised clustering via voluntary tagging, like a hashtag on twitter, and couple that with geolocation. From there you have the classification of information and can track that to how the user interacts. I agree it COULD result in mis-classification, but there’s a lot being done quite well with unsupervised clustering that’s proving effective. Users on at least some services such as twitter are already providing tagging of other users, and validating which items fall in a category via tagging. By expanding the classification to just beyond California, to California and Traveling you could clearly identify that as an area of interest, thereby eliminating things only pertaining to California and only pertaining to Traveling. Essentially any filtering you could do would be some improvement, and how effective you could be would be more restricted by the amount you wanted to see in your feed as much as it would on how effective your filtering was. Only for the users that greatly limited what they wanted in a feed would you miss out. Since you’re already opting in to relevant users, instead of on the entire corpus of twitter you have to worry less about the overload of information of what may occur for conflicting classifications.

I think your initial question really becomes the driver for then being the input to filtering the stream. You begin with unsupervised clustering via voluntary tagging, like a hashtag on twitter, and couple that with geolocation. From there you have the classification of information and can track that to how the user interacts.

]]> By: Matt http://www.craigekerstiens.com/who-will-filter-the-stream-first/comment-page-1/#comment-25530 Matt Tue, 26 Jan 2010 20:26:49 +0000 http://www.craigekerstiens.com/?p=160#comment-25530 To answer your last question: because products have a defined set of meta-data and context in which that data is useful. They have years upon years worth of behavioral patterns stored to determine similar interest among items. The problem with the information on Facebook and other social sites is that the information isn't nearly as easily placed into context using artificial intelligence. The amount of AI work to do something like the example you listed with "California" could easily result into you seeing someone is currently enjoying a Tupac and Dr. Dre track - something you aren't very likely interested in. I think the better question becomes: how do you convert a user's data into actual information? The first obvious step to me is meta-data, even if it means having the user voluntarily qualify and add extra information to their shared data. Figuring out this meta-data on a 140 character sentence seems impossible without some type of user input. I think we already have a lot of tools to help us make a more educated guess: geolocation, "tagging" of other users, etc., but we still have a long way to go. To answer your last question: because products have a defined set of meta-data and context in which that data is useful. They have years upon years worth of behavioral patterns stored to determine similar interest among items.

The problem with the information on Facebook and other social sites is that the information isn’t nearly as easily placed into context using artificial intelligence. The amount of AI work to do something like the example you listed with “California” could easily result into you seeing someone is currently enjoying a Tupac and Dr. Dre track – something you aren’t very likely interested in.

I think the better question becomes: how do you convert a user’s data into actual information? The first obvious step to me is meta-data, even if it means having the user voluntarily qualify and add extra information to their shared data. Figuring out this meta-data on a 140 character sentence seems impossible without some type of user input. I think we already have a lot of tools to help us make a more educated guess: geolocation, “tagging” of other users, etc., but we still have a long way to go.

]]>