Design and implementation of the consumer buying process for mobile agent-based Internet commerce system (MAGICS)

Abstract

With the advent of software agent technologies, there has been considerable interest in the development of agent-based e-commerce systems. Agents can make a computer system more effective because they can perform tasks autonomously. To complement current Web-based Internet commerce systems and particularly to support consumer-oriented e-commerce, we have developed a mobile agent-based e-commerce system called Business-to-Consumer (B2C) Mobile AGent-based Internet Commerce System (MAGICS). This project focuses on using B2C MAGICS to facilitate the consumer buying process. Consumers can use B2C MAGICS through a proxy server via a Web/WAP interface. In accordance with user requirements, agents are sent to obtain responses from sellers. The best seller is chosen based on these responses and an agent is sent to carry out the transaction. To complete these tasks, the agents need to exchange many messages. In this thesis, we develop an Extensible Markup Language (XML) scheme for inter-agent communications in which agents communicate through MAGICS messages which are specified by an XML schema. A MAGICS message consists of three kinds of information: basic, functional, and additional. It is possible to use this framework to define messages that support the agent-based consumer buying process. We also investigate a product-comparison problem. When an agent visits a list of shops sequentially to perform search and evaluation, the decision whether to buy the item or continue shopping will entail comparisons. We consider two kinds of comparison, price and multi-attribute. In the case of price comparison, we formulate two Markov-decision-based price comparison models. This model uses a backward induction algorithm which determines the optimal decision policy. This allows the agent to buy a product at the minimum expected cost, including the traveling cost. The model is analyzed using a normal price distribution and real price information and simulations are conducted to validate the analytical results. For the case of multi-attribute comparison, we propose a Fuzzy Markov Decision Process (fMDP) for Multi-attribute Product Comparison, which is extended from the price comparison models. The fMDP model handles multiple attributes evaluation using a Fuzzy Logic algorithm. This produces an optimal decision policy that an agent can use to buy the most suitable product. Analytical results are presented to study the behavior of the system with different parameters. Furthermore, experiments are carried out to compare the performance of the fMDP model and the performance of one of the price comparison models.