摘要
Shill bidding has become a serious issue for innocent bidders with the growing popularity of online auctions. In this paper, we study the relationship between final prices of online auctions and shill activities. We conduct experiments on real auction data from eBay to examine the hypotheses that state how the difference between final auction price and expected auction price implies shill bidding. In the experiments, a neural network based approach is used to learn the expected auction price. In particular, we trained the Large Memory Storage and Retrieval (LAMSTAR) Neural Network based on features extracted from item descriptions, listings and other auction properties. The likelihood of shill bidding is determined by a previously proposed shill certification technique based on Dempster-Shafer theory. By employing the chi-square test of independence and logistic regression, the experimental results indicate that a higher-than-expected final auction price might be used as direct evidence to distinguish likely shill-infected auctions from trustworthy auctions, allowing for more focused evaluation of shill-suspected auctions. As such, this work contributes to providing a feasible way to identify suspicious auctions that may contain shill biddings. It may also help to develop trustworthy auction houses with shill detection services that can protect honest bidders and benefit the auction markets in both the short-term and long term.