An Empirical Study on the Relationship Between Final Auction Price and Shilling Activity in Online Auctions

In this paper, we are interested in the relationship between final prices of online auctions and possible shill activities during those auctions. We conduct experiments on real auction data from eBay to exam the hypotheses that (1) A lower-than-expected final auction price indicates that shill bidding was less likely to occur in that auction; and (2) A higher-than-expected auction final price indicates possible shill bidding. In the experiments, a neural network approach is used to learn the expected auction price. In particular, we used the LArge Memory Storage and Retrieval (LAMSTAR) Neural Network based on features extracted from item descriptions, listings and other auction features. The likelihood of shill bidding is determined by a previously proposed Dempster-Shafer theory based shill certification technique. The experimental results show that both a lower-than-expected final auction price and a higherthan-expected final auction price can be used as evidence to distinguish trustworthy auctions from likely shillinfected auctions, allowing for more focused evaluation of those shill-suspected auctions.