Abstract
M.Sc.
This dissertation investigates how auctioneer agents can maximise the revenue of an auction. Auctions are an effective solution to agent negotiation because of their simplicity. They are therefore the most widely used approach to agent negotiation. A review of auction theory proves that auction revenue is influenced by factors such as the auction format and the auction parameters. The optimal auction format and parameters are dependent on the bidders and the auction environment. A study of internet auctions confirms the findings of auction theory regarding the importance of the format as well as the parameters in an auction. There has been significant research on the strategies bidding agents can adopt in order to maximise their utility. There is little research, however, on the strategies auctioneer agents can adopt to maximise auction revenue. In this study, a survey is done of the related auctioneer agents in research that maximise auction revenue. The shortcomings of these are identified. The aim of this research is to develop an adaptive auctioneer agent that can maximise the revenue of an auction. The proposed solution is the revenue maximising adaptive auctioneer (RMAA) agent. The RMAA agent consists of three components: the auction manager, the auction format selector (AFS) and the auction parameter selector (APS). The AFS employs a customised zeroth-level classifier system to select the auction format based on the auction environment. The APS employs reinforcement learning combined with a local hill climbing algorithm to learn what the most profitable values are for auction parameters, such as the reserve price and the bid increment. The auction manager automates the auction and coordinates the AFS and APS. It implements the four traditional auction formats, namely the English auction, the Dutch auction, the Vickrey auction and the first-price sealed-bid auction. Experiments conducted to evaluate the effectiveness of the RMAA agent prove that this agent can consistently generate more revenue than a non-adaptive auctioneer agent. The contribution of this research is an auctioneer agent that adapts both the auction format and the auction parameters according to the environment in order to maximise the auction revenue.