TETRA: TExtual TRust Analyzer for a Gricean approach to social networks

Social ties in human relationships are often based on trust between peers. Depending on the context, trust can assume different forms, and can be computed and quantified in different ways. In this work, we introduce Tetra, a framework based on the theory of cooperative principle by the linguistic Paul Grice, employing state-of-the-art NLP techniques to assign three different trust scores to a sentence, focusing on relation between sentences (Relation), information density (Quantity), and politeness (Manner). Furthermore, we employ the framework to analyze a network of Reddit users in order to identify how trust scores can be leveraged to get a better insight into human relationships, assuming that the trust scores computed by Tetra can be applied to the network’s edges, or averaged to assign a score to the network’s nodes. Our experiments showed that trust scores computed by Tetra can be employed to cluster the network’s nodes, can successfully validate another independent network trust model and can be used to gather interesting insights in the context of Social Balance Theory.