The web has been made much more interesting with social network sites. Its evolution from a technology platform to a social milieu has had profound impact on how people connect with each other. Sites like del.icio.us, Flickr, Facebook, etc., have become an essential part of life. Well, at least for one of us :-)
One of the fundamental challenges facing many of social network sites is how to help the users explore the site content effectively and efficiently. With the rapid growth of those sites, the problem is only becoming more important.
So how is searching information on social network sites different from searching on the web? First, it's not really search. Users on social network are often looking for something interesting within a given topic, instead of searching for any specific page or item. This process requires an approach that is very much a blend of recommendation, search, and sometimes database style querying. What's more important, the recommendation strategies being employed need to be flexible such that it can respond to changing user needs. Second, the people and community are important. The information about a user within a social network site is so much richer that opportunities for improving the user's experience are plenty. Third, the information needs are rich. I don't know for you guys, but I frequently find myself interested in knowing more about users on del.icio.us and Flickr. All these things have led us to start the Royal Jelly project, which is aimed to provide a flexible recommendation platform for helping users explore contents on social network sites.
There are many interesting research ideas and projects (we call them our Bees) being discussed within the Royal Jelly framework. Today, we are introducing you to one of the Bees: Garçon, a prototype implementation of Royal Jelly on top of Yahoo!'s social bookmarking site del.icio.us. Garçon analyzes a user's friendship network and behavioral profile to make recommendations to the user on what they might find interesting. There are several interesting features. First, Garçon recommends more than just the bookmarks, but also people and topics (as represented by tags). Second, for each recommendation, a brief explanation is provided so that you can understand why the recommendation is made. Finally, the challenge of analyzing the amount of user and bookmark information at the scale of del.icio.us is being addressed using Hadoop and Pig.
There are many features that we will gradually add in the future. Those high on the priority list include allowing user to declare the recommendation strategy they would like to use (currently the strategies can only be declared by a system admin), introducing guided recommendation where a search or query condition is also provided by the user, and allowing feedback on recommendations. We are also extremely interested in suggestions and thoughts from you!
For those of you who are concerned about privacy: recommendations in Garçon are based only on "public" del.icio.us bookmarks. We do not and will not attempt to leverage any private activities in creating recommendations.
Finally, Garçon is a team effort! Jacob Leatherman, George Levchenko, Pras Sarkar, and Tejaswi Kasturi at Yahoo! Research certainly did more than the two of us :-) And Joshua and Stephen from the del.icio.us team provided many help from the very beginning.
Tell us what you think, leave a comment: