Trust Mining and analysis in complex systems
A complex system, as a collection of loosely coupled interacting components, can group and create functioning units together. Complex systems have become a powerful framework for describing, analysing, modelling systems in nature and society. Trust among components, established by considering past interactions, represents a subjective expectation which a component has about another’s future behaviour to perform given activities dependably, securely, and reliably. Hence, trust is essential to effectively reduce the perceived risks of transactions and guide future interactions. It is applied to quantify the performance of both individual component behaviours and the correlations among interdependent components in a complex system.
With regard to certain challenges in the current complex system research, this thesis deeply investigates trust relationships among components within two different types of complex systems, i.e., the collaborative complex system and the preference system, and proposes three trust estimation approaches. Firstly, collaborative complex systems consist of loosely coupled autonomous and adaptive components. In order to address complicated problems which usually require multiple skills and functions, components are grouped as composite teams and collaborate by providing different knowledge, resource and skill. Two types of team formation strategies for collaborative complex systems are proposed for scenarios of team formation without predefined workflow structures, and team formation with predefined workflow structures, respectively. Hence, the Correlated Contribution trust evaluation model is proposed to explore the compositional trust through considering correlations and dependencies among both skills required by tasks and individual components within collaborative composite teams. Furthermore, we propose an automatic approach, i.e., the Same Edge Contribution trust evaluation model, to estimate the trustworthiness of proposed candidate composite teams by analysing historical provenance graphs which are adopted to capture pre- defined workflow structures. Finally, preference systems mainly focus on the entities with similar preferences and group them into various communities. However, in the real world, a particular entity usually places its trust differently from other social entities, because of their multi-faceted interests and preferences. In this thesis, a Community- Based trust estimation approach is proposed to explore the similarity of criteria or preference among entities within the same community in relation to a certain context. It automatically infers trust relationships among entities from previous entity-generated feedback, and predict a particular entity’s potential feedback for items which the entity does not have previous experience with.