Ανάπτυξη μηχανισμών πλειοδοσίας για πράκτορες ηλεκτρονικών δημοπρασιών με χρήση ιστορικών δεδομένων

Abstract

The objective of this thesis is the design and development of bidding mechanisms for intelligent software agents that participate in different types of online auctions. In the auction environment, agents try to maximize their revenue, operating in a competitive manner against their rivals. Each agent may represent one or more clients and must make several inter-related decisions in order to obtain the best possible profit. For example, it must decide which is the most profitable auction, when to bid in that auction, as well as what the best price of the submitted bid should be. One important aspect of an agent is its ability to predict the closing price of an auction. This is especially useful for the calculation of the most profitable auction and the deployment of efficient bidding mechanisms based on the prediction of the closing price. Within the context of this dissertation several bidding mechanism have been developed. These take into account historical data produced in the auction environment. Specifically, two types of historical data are taken into consideration: a) intermediate data produced while the auction is still in progress and b) data from past auctions. We have developed two classes of forecasting techniques for each one of the aforementioned data types. These are known as a) short-term and b) long-term forecasting, respectively. Different auction types have been studied and different bidding mechanisms have been developed for each one of them. All mechanisms implement some form of price forecasting based on historical data. To evaluate the performance of a bidding agent, a benchmarking platform is used, which typically calculates the total profit gained in a predefined number of auctions that the agent participates in. In this thesis we have both developed appropriate benchmarking platforms, when necessary, and used already developed and widely adopted benchmarking environments. One such platform is provided by the international Trading Agent Competition. The use of these platforms allow the comparison of our techniques against others, reported in the literature.