Analyzing Trust-Based Mixing Patterns in Signed Networks

Abstract

In some online social media such as Slashdot, actors are allowed to explicitly show their trust or distrust towards each other. Such a network, called a signed network, contains positive and negative edges. Traditional notions of assortativity and disassortativity are not sufficient to study the mixing patterns of connections between actors in a signed network, owing to the presence of negative edges. Towards this end, we propose two additional notions of mixing due to negative edges – anti-assortativity and anti-disassortativity – which pertain to the show of distrust towards “similar” nodes and “dissimilar” nodes respectively. We classify nodes based on a local measure of their trustworthiness, rather than based on in-degrees, in order to study mixing patterns. We also use some simple techniques to quantify a node’s bias towards assortativity, disassortativity, anti-assortativity and anti-disassortativity in a signed network. Our experiments with the Slashdot Zoo network suggest that: (i) “low-trust” nodes show varied forms of mixing – reasonable assortativity, high disassortativity, slight anti-assortativity and slight anti-disassortativity, and (ii) “high-trust” nodes mix highly assortatively while showing very little disassortativity, anti-assortativity or anti-disassortativity.