The Strength of Negative Opinions.

We investigate signed social networks, in which users are connected via directional signed links indicating their opinions on each other. Predicting the sign of such links is a crucial task for many real world applications like recommendation systems. Based on the premise that like-minded users tend to influence each other more than others, we present a logistic regression classifier built on evidence drawn from the users' ego-networks. The main focus of this work is to examine and compare the relative strength of positive and negative opinions investigating to what extent each type of link affects the overall prediction accuracy. We evaluate our approach through a thorough experimental study that comprises three large-scale real-world datasets.