Detecting leaders to alleviate latency in recommender systems

Abstract

The exponential increasing of information on the Web and information retrieval systems engendered a heightened need for content personalization. Recommender systems are widely used for this purpose. Collaborative Filtering (CF) is the most popular recommendation technique. However, CF systems are very dependent on the availability of ratings to model relationships between users and generate accurate predictions. Thus, no recommendation can be computed for newly incorporated items. This paper proposes an original way to alleviate the latency problem by harnessing behavioral leaders in the context of a behavioral network. In this network, users are linked when they have a similar navigational behavior. We present an algorithm that aims at detecting behavioral leaders based on their connectivity and their potentiality of prediction. These leaders represent the entry nodes that the recommender system targets so as to predict the preferences of their neighbors about new items. This approach is evaluated in terms of precision using a real usage dataset. The results of the experimentation show that our approach not only solves the latency problem, it also leads to a precision higher than standard CF. © 2010 Springer-Verlag Berlin Heidelberg.