WhatsUp is a recommendation system aimed to distribute instant news in a large scale dynamic system with no central bottleneck, single point of failure or censorship authority. Users express their opinions about the news they see by pushing a button (like-dislike) or simply ignoring them. The users expect to subsequently receive more interesting news, following a collaborative filtering scheme, based on what news have interested users with similar taste without explicitely subscribe to topics of interest. Underlying WhatsUp are two distributed protocols. The first, Wup, dynamically clusters active users with similar tastes using a multi-interest cosine similarity metric accounting for emerging interests. The second, Beep, is used to disseminate news from a user to a randomly selected set of users with a bias towards those with similar interests. The size of that set itself depends on the interest expressed by the user about the news it read. WhatsUp has been implemented and extensively tested using traces involving thousands users from system Digg. We report on its efficiency to deliver relevant news to interested users while limiting spam.
- Demonstration of WhatsUp will be held during the 1st Gossple Workshop
Video showing the cluster formation and dissemination in WhatsUp:
- Pictures of the demonstration given at P2P 2010 (IEEE International Conference on Peer-to-Peer Computing), August 2010.