Preference and Diversity-based Ranking in Network-Centric Information Management Systems

Diversity in Publish/Subscribe Systems

With the explosion of the amount of information that is currently available online, publish/subscribe systems offer an attractive alternative to searching by providing a proactive model of information supply. In such systems, users express their interest in specific pieces of data (or events) via subscriptions. Then, they are notified whenever some other user generates (or publishes) an event that matches one of their subscriptions. Typically, all subscriptions are considered equally important and users are notified whenever a published event matches any of their subscriptions.

Basic Pub/Sub

A basic publish/subscribe system.

However, getting notified about all matching events may lead to overwhelming the users with huge amounts of notifications, thus hurting the acceptability of publish/subscribe systems. To control the rate of notifications received by the subscribers, it would be useful to allow them to rank the importance or relevance of events. Then, they would only receive notifications for the most important or relevant among them. For example, take a user Addison that generally likes drama movies but prefers drama movies directed by T. Burton to drama movies directed by S. Spielberg. Ideally, Addison would like to receive notifications about S. Spielberg drama movies only if there are no, or not enough, notifications about T. Burton drama movies.

However, the top-k events are often very similar to each other. Besides pure accuracy achieved by matching the criteria set by the users, diversification, i.e., recommending items that differ from each other, has been shown to increase user satisfaction. For instance, our user Addison would probably like to receive information about different drama movies by T. Burton as well as a couple of S. Spielberg's drama movies once in a while. To this end, we consider adjusting the top-k computation to take also into account the diversity of the delivered events. To achieve this, we consider both the importance of each event as specified by the user preferences as well as its diversity from other top-ranked events.

Related Publications:

  • Marina Drosou, Kostas Stefanidis and Evaggelia Pitoura, Preference-Aware Publish/Subscribe Delivery with Diversity, in Proc. of the 3rd ACM International Conference on Distributed Event-Based Systems (DEBS 2009), July 6-9, 2009, Nashville, TN, USA (Also presented at the 8th Hellenic Data Management Symposium (HDMS), August 31 - September 1, 2009, Athens, Greece)
  • Marina Drosou, Ranked Publish/Subscribe Delivery, PhD Workshop, in conjunction with the DEBS 2009 Conference, July 6, 2009, Nashville, TN, USA
  • Marina Drosou, Evaggelia Pitoura and Kostas Stefanidis, Preferential Publish/Subscribe, in Proc. of the 2nd International Workshop on Personalized Access, Profile Management and Context Awareness: Databases (PersDB 2008), in conjunction with the VLDB 2008 Conference, August 23, 2008, Auckland, New Zealand