A Trust Management Framework for Decision Support Systems

In the era of information explosion, it is critical to develop a framework which can extract useful information and help people to make educated decisions. In our lives, whether we are aware of it, trust has turned out to be very helpful for us to make decisions. At the same time, cognitive trust, especially in large systems, such as Facebook, Twitter, and so on, needs support from computer systems. Therefore, we need a framework that can effectively, but also intuitively, let people express their trust, and enable the system to automatically and securely summarize the massive amounts of trust information, so that a user of the system can make educated decisions, or at least not blind decisions. Inspired by the similarities between human trust and physical measurements, this dissertation proposes a measurement theory based trust management framework. It consists of three phases: trust modeling, trust inference, and decision making. Instead of proposing specic trust inference formulas, this dissertation proposes a fundamental framework which is flexible and can be adapted by many different inference formulas. Validation experiments are done on two data sets: the Epinions.com data set and the Twitter data set. This dissertation also adapts the measurement theory based trust management framework for two decision support applications. In the first application, the real stock market data is used as ground truth for the measurement theory based trust management framework. Basically, the correlation between the sentiment expressed on Twitter and stock market data is measured. Compared with existing works which do not differentiate tweets\u27 authors, this dissertation analyzes trust among stock investors on Twitter and uses the trust network to differentiate tweets\u27 authors. The results show that by using the measurement theory based trust framework, Twitter sentiment valence is able to reflect abnormal stock returns better than treating all the authors as equally important or weighting them by their number of followers. In the second application, the measurement theory based trust management framework is used to help to detect and prevent from being attacked in cloud computing scenarios. In this application, each single flow is treated as a measurement. The simulation results show that the measurement theory based trust management framework is able to provide guidance for cloud administrators and customers to make decisions, e.g. migrating tasks from suspect nodes to trustworthy nodes, dynamically allocating resources according to trust information, and managing the trade-off between the degree of redundancy and the cost of resources