Dec. 8, 2006 at 8:00 AM ET
EBay users worried about fraud, take heart: Researchers at Carnegie Mellon University say they've invented data mining techniques that can detect online auction fraud before it happens. By looking at trading patterns among thousands of eBay sellers, the researchers say they can identify likely con artists and warn off potential bidders.
“We want to help people detect potential fraud before the fraud occurs,” said research associate Duen Horng Chau, who developed NetProbe with professor Christos Faloutsos, undergraduate student Samuel Wang and graduate student Shashank Pandit.
EBay sellers gain their credibility largely through the use of an extensive user-generated feedback system. Buyers give good marks to sellers who deliver goods as promised, and generally sellers with high feedback ratings are safer to purchase from. But con artists long ago found ways to "game" the feedback system, either by creating fake eBay buyers to artificially inflate their positive rating or by working with other criminal associates.
Carnegie Mellon's system, called NetProbe, examines connections between buyers and sellers to look for patterns of artificial feedback. By drawing connections between groups of eBay users, the software can make associations between otherwise anonymous eBay sellers and known criminals, for example.
"Essentially, we are trying to find fake users," Faloutsos said.
These associations, which might not be apparent to the naked eye, can then be used to warn potential buyers, Faloutsos said.
He said the system works because it is able to scan several layers deep into connections between eBay users. "For example, we are looking at the people who traded with the people who traded with me," Faloutsos said.
In its most recent test, Carnegie Mellon researchers loaded the full transaction history of 70,000 eBay users -- 1 million trades in all -- into a database. Running the data through NetProbe, the researchers were able to correctly pick out 10 previously identified criminals. They also identified more than a dozen probable fraudsters and several dozen apparent accomplices, the university said.
Larger tests are planned, and so is a commercial product, though school officials said they didn’t know if NetProbe would be offered directly to consumers. Theoretically, though, potential bidders could run a one-click fraud test on a seller and get back a score indicating the likelihood that the seller was a criminal, Chau said.
Scoring systems common online
That would be a bit of Internet turnabout. Similar scoring systems are used to analyze risk levels in many Internet-based transactions, but they are generally used by sellers to analyze buyers. Most e-commerce sites use fraud scoring products to detect the likelihood that a buyer is using a stolen credit card to purchase merchandise, for example.
The NetProbe software can actually be used to predict eBay crime because it takes some time for criminals to artificially inflate feedback scores, the researchers say. Even the shortest eBay auction takes one day.
“Real fraudsters who want to make fake reputations, they still need some time to build those,” Faloutsos said. By detecting the creation of fake reputations in real time, fraud can be stopped before it starts, he said.
EBay had not been consulted in development of the software, Faloutsos said, and the company did not immediately return calls seeking comment.
Fake feedback not the biggest problem
EBay fraud expert Rosalinda Baldwin, who runs The Auction Guild Web site, welcomed the research project but said she was skeptical of its widespread usefulness.
Manipulated feedback scores, while a problem for eBay, are not the site's chief source of fraud, she said. Most eBay criminals use hijacked account to commit fraud, allowing them to impersonate legitimate, well-respected eBay members. NetProbe's relationship testing would be powerless to find such fraud, she said.
Consumers can also do their own relationship analysis using eBay's site, she said. In fact, no one should ever bid on an item and not click through a seller's feedback contributors to see if anything looks suspicious, she said.
She said NetProbe would be useful at stopping another common eBay problem – shill bidding, in which sellers artificially raise prices by bidding on their own items using different accounts.
"It would be of great use to a buyer to be able to run a search on the seller and see if there appears to be shill bidding activity on the sellers account," she said.
Warning: Strange 1-cent auctions
On the subject of bidding, eBay users should know about the increased prevalence of “feedback groups,” designed to artificially inflate scores through a series of fake low-priced auctions – many listing items for sale a 1 cent each. The groups buy items from each other essentially for free, and give positive feedback ratings to each other. Baldwin says this is common among cooperating criminals. But sometimes, unwitting accomplices are taken along for the ride.
That’s what happened to Larry Harris, an economics professor at the University of Southern California and former chief economist at the Securities and Exchange Commission. Harris’ eBay account was recently hacked somehow, and in one day his impersonator won 13 auctions by bidding 1 cent or 5 cents for items that were essentially worthless, such as an electronic book. Positive feedback had already been given by the time Harris noticed the action, changed his password and regained control of his account.
“When I looked at sellers, I saw thousands of feedback,” Harris said. “People are clearly using some kind of automated system to do these frauds.”
Baldwin said the case should serve as a warning to eBay users that they need to beware of sellers who have gained their reputations from sale of such low-priced items.