Trading without Explicit Ontologies Michael Schroeder, Julie McCann, and Daniel Haynes City University, London msch,jam @soi.city.ac.uk Abstract. Classical approaches to traders in middleware rely on a common lan- guage of server, clients, and traders to understand each other. In these systems, pre-defined ontologies play a crucial role. But when dealing with large-scale, open systems such ontologies are no longer available. To cope with this problem, we have developed a radically different approach to trading. Rather than relying 100% on a trader, we assume that traders provide only rough matches and that clients need to make intelligent choices to find a more suited service. To this end, we introduce the notion of trust, which evolves with the client’s experience. We implement a simulation of this trust-based trading system and run several test scenarios investigating the use of history and dynamic trust to discover the more suited services. Our analysis and simulation indicate that intelligent clients and rough traders may considerably extend the scope of trading towards large-scale, open distributed systems. 1 Introduction A trader is a middleware component currently used to match requests for services to services. They are typically used in relatively closed environments where a single de- fi ned language and set of semantics are used to describe requests and services. Many researchers are beginning to see a broader use of trading within a more open distributed system, such as the Web. In this environment it is expected that trading systems need to be able to carry out more fuzzy mappings between requests and services as the se- mantics and languages are no longer standard. Further, systems which request services (clients) need to be able to acquire a level of trust in both the traders and the services when deciding to use them in future. The client must be able to break away from a service that it notices is deteriorating, so as to obtain a better one. Further, it is also en- visaged that future services, and in fact the trading service itself, will charge for usage. Current trading systems are unable to provide this level of service; in this paper we provide an alternative which does. This system stores a history of quality of services to establish its trust, carry out fuzzy matching of services while providing the ability to dynamically rebind from one service to another. The perception of trust must also take into account that a service will cost and this cost may fluctuate also. To this end, we propose a new trading system and have built a simulation of its behaviour so as to study how metrics such as trust and costs affect client/trader selection. The structure of this paper is as follows. Firstly, we introduce the concept of middleware trading In Proc. of Agent-mediated E-commerce, Barcelona, Sept. 1999, Springer-Verlag