Speaker: Li Ding
Start: Monday, April 05, 2004, 11:00AM
End: Monday, April 05, 2004, 12:00PM
Location: ITE 325B
Abstract: The growth in knowledge sharing enabled by the (semantic) web made trust an increasingly critical issue. A trust network refers to a network of agents connected by directed trust relations with Boolean or numerical values. We argue that it is more personalized, flexible and scalable than traditional approaches such as public reputation systems and collaborative filtering. Some researchers have recently suggested a graph interpretation of the trust network, and proposed simple algorithms for propagating trust. However, it is not clear that trust is completely transitive or even static. This paper systematically characterizes trust network inference with three components: a data model, a computational model and an evaluation model. The data model clarifies the context (input, constraint and output) of trust network inference for knowledge sharing. It also elaborates trust network representation and different types of trust. The computation model reviews existing, graph theory based trust network computation models, and proposes a new model that treats trust as an emergent property. The model supports both trust evolution and trust propagation. The evaluation model describes evaluation metrics, as well as methods to generate test data. We also design and implement an agent based trust network evaluation framework.