Recommender Systems have been proved effective to apply knowledge discovery techniques to the problem of providing personalized recommendations for products and services to users during a live interaction. Clustering and trust-based models have been proposed as improvements to existing recommendations algorithms. While they can be helpful when used separately, the combination of using social trust data for clustering users has not been studied adequately so far.
In this work we explore well established clustering schemes, such as k-means as well as new ones, and we demonstrate the advantages in performance from the use of social-oriented information for clustering. The strong points of this approach include the lower computational cost, the higher resistance to manipulation of user preferences, as well as the fact that such information can be provided by the social networks.
Georgios Pitsilis received his B.Eng in Engineering in 1991 and B.Eng in Informatics in 1999, both from the Technological Educational Institution of Athens, Greece, his MSc. degree from Oxford Brookes University-UK in 2000, and PhD degree from Newcastle University-UK in 2007. Now he is a postdoc researcher at the University of Luxembourg. Prior to this position he has been working as research scientist at Newcastle University UK in an e-Science project, and later as postdoc fellow at NTNU, Norway.
