(A) flexible conversation model for multi-agent systems in dynamic electronic market environments

Abstract

In an electronic market (e-market) environment, there can be many changes in products and transaction rules, and, therefore, information systems need to be designed in a flexible way to meet the changing requirements. Multi-agent systems (MAS) have been rigorously researched to enable autonomous transactions in the e-market environment. However, they have been interoperable only in a limited way until now, requiring expensive costs and time for adaptation. In agent-based electronic markets, the changes accompanied by new trading partners or new products are reflected in the conversation policies of agents. A conversation policy is a predefined sequence of agent messages that constrain and guide the exchange of messages. Conversation policies are usually defined for specific types of agent transactions such as auction, brokering, or negotiation. Because existing agent systems allow only a fixed set of hard-coded conversation policies, adaptation to new transaction methods requires significant modification and re-implementation of agent systems, which reduces the overall efficiency of electronic markets. In order to address this problem, this dissertation suggests a flexible conversation model (FCM) that enables agent systems to acquire and use new conversation policies in an on-demand way. By using FCM, agent systems can adapt to new forms of transactions by acquiring new conversation policies when changes occur in the electronic market. FCM consists of an interpretable conversation policy model for defining instances of exchanged conversation polices, a handshaking procedure for exchanging new conversation policies between agents in a mutually-agreed way, and a mechanism for performing transactions with newly acquired conversation policies. In order to effectively utilize FCM, an ontology meta-model that can facilitate the development and validation of domain ontologies is also proposed. The ontology meta-model called TOO divides domain ontologies into contexts...