Trust in ubiquitous computing is about finding trustworthy partners for risky interactions in presence of uncertainty about identity, motivation, and goals of the potential interactions partners. In this paper, we present new approaches for estimating the trustworthiness of entities and for filtering and weighting recommendations, which we integrate in our trust model, called CertainTrust. We evaluate the robustness of our trust model using an canonical set of population mixes based on a classification of typical entity behaviors. The simulation is based on user traces collected in the Reality Mining project. The evaluation shows the applicability of our trust model to collaboration in opportunistic networks and its advantages in comparison to a distributed variant of the Beta Reputation System.
展开▼
