Social interaction propensity model using information entropy

This paper introduces a novel user model, social interaction propensity model, for computing similarity of mobile phone users. Traditional studies exploit the usage history to represent the users by their behavioral patterns. This representation model requires prohibitive costs for dealing with the high-dimensional space that contains the usage patterns according to various contextual features. To alleviate the high-dimensionality, we propose a user model that is represented no longer with explicit usage patterns but only with its distribution uniformity to reduce the space. For evaluation, we developed a life-logger application to gather the real data from users. The evaluation result indicates that the user space is reduced linearly with the number of features without losing the precision of computing user similarity.