Abstract

Current approaches to social search are generally based on the friends’ opinion and collaborative filtering paradigms, which have limited effectiveness in terms of both accuracy and coverage of the information that may be relevant to a user. We propose a new kind of social search signal, based on the idea of prediction extraction, which is aimed at improving the quality of information available to current methods, as well as introducing new kinds of search methods.