Trust is a key concept in social networks, reflecting credibility and reliability for a multitude of participants and online data. Nevertheless, the majority of such networks lack explicit trust feedback. This motivates a mechanism to predict and manage trust relations automatically. We extract in an unsupervised manner local trust relationships between pairs of users from social networks derived from Twitter. We take into account factors of measurable influence between users, the impact of the structural topology of users in the network and the valence (sentiment) associated with the languagebased information shared by network members. We evaluate our user trust rankings over other members of the network against a metric of ground truth for both social media data and a non-social media dataset, and analyze how the inclusion of valence lends robustness and stability to our model of trust. Knowledge of trustworthy citizens in social networks is quite advantageous in accurately assessing the credibility of the information they provide via social media, for the purpose of emergency response and recovery efforts during a disaster or a catastrophic event.