Cooperative team formation using distributed decomposition knowledgeA thesis submitted in August 2010 by Martin Carpenter to the University ofManchester for the degree of Doctor of Philosophy. In recent years, the problem of automating the formation of Virtual Organisations (VO) has risen to prominence. Work in this area has typically considered the process of VO formation to be a centralised process driven by a company with responsibility for the business opportunity.Such systems use two main stages: first they decompose the business opportunity into a set of roles and then select suppliers for each role by matching their advertised capability against criteria supplied by the user. Both stages require that the company driving the VO formation process has access to considerable amounts of centralised knowledge.In contrast, this thesis considers virtual organisations as forming by combining the cooperative contributions within a group of organisations. It is shown that, within this context, both the knowledge required to facilitate and the control within the virtual organisation formation process are naturally distributed. In particular companies are free to vary their level of commitment to given projects and so only they have detailed knowledge of their capabilities. Supporting VO formation within this context requires a novel approach capable of utilising this distributed information. The primary contribution of this thesis is to provide such a novel approach to supporting virtual organisation formation. This approach builds on the traditions of blackboard and multi-agent systems. It allows virtual organisation formation to be driven by the accumulation of voluntary contributions from the prospective members of the virtual organisation. The principle focus of the system is on identifying candidate virtual organisations, and it does not offer automated support for such aspects as the creation of contracts. Crucially this system works with the distributed knowledge encountered in the chosen problem domain. The following technical contributions shape the general approach into a detailed system: (a) the representation of company's capabilities, (b) an algorithm for combining those capabilities and (c) mechanisms enabling intelligent agents representing the companies to produce candidate virtual organisations. The proposed system is evaluated in three ways - its technical feasibility is demonstrated through the implementation of a testbed prototype, a theoretical discussion of the systems performance is given and finally its potential benefits are shown in a reasoned case study.