Predicting Trust Relations Within a Social Network: A Case Study on Emergency Response

Trust is a fundamental construct underpinning modern society and the social exchanges it contains. The rise of Web 2.0 technologies and the increased use of online social networks, promotes the study of trust among users. Drawing on social and psychological theory, we detect pairwise and global trust relations between users in the context of emergent real-world crisis scenarios. In such situations and scale, seeking explicit pairwise trust assessments between users is impractical. Instead, in an unsupervised manner we integrate the implicit factors of social influence exerted by each user over the network, the underlying network structural topology and the affective valence expressed by the users in the textual content they communicate. A key finding is the importance of modeling influence and affective valence in such exchanges and their role in detecting stable trust relationships. We extensively evaluate these ideas and demonstrate significant gains over competitive baselines across multiple datasets drawn from both crisis and non-crisis scenarios, including those with normative ground truth.