Abstract
The current state of electronic commerce in tourism shows that it has become an increasingly complicated task for travellers to locate and integrate disparate information as a result of the rapid growth in the number of online travel sites. Therefore, new means of automating the searching and decision-making tasks are needed. A review of current literature shows that software agents are deemed to be highly suitable for delivering solutions to these problems. However, agents have failed to penetrate the electronic marketplace so far. An analysis of the reason for this failure has led the author to conclude that a new type of architecture is required, allowing a simple and useful first wave product to accelerate the penetration of agents. For this purpose, a proof-of-concept multi-agent prototype - Personal Travel Assistant (PTA) was developed. Firstly, user requirements were compared against what existing network and agent technologies could deliver. Then, a number of obstacles were identified that were used as guidelines to derive the prototype architecture. To overcome the main obstacles in the design, PTA used existing HTTP servers to tackle the interoperability problem and keep development costs low. A multi-agent collaborative learning strategy was designed to speed up knowledge acquisition by transferring and adapting rules encoded in the Java language. The construction of PTA goes to prove that an open multi-agent system could be deployed in a short time by standardising a small but adaptable set of communication protocols instead of going through a complex and lengthy standardisation process. Also, PTA's structure enables fully distributed computing thus minimising the necessary changes in existing hardware and software infrastructure. The major contribution of PTA to this research area is that its architecture is unique. It is hoped that it will lay the first step on the roadmap that would lead the evolution of agents into the next stage of development.