Software that could help catch fraudsters hoping to deceive other users on internet auction sites has been developed by US researchers. It analyses a user鈥檚 transaction history to determine whether they might be deliberately trying to build a false 鈥渞eputation鈥� for themselves.
Auction site like eBay rely on a system of trust to prevent abuse. Buyers and sellers provide feedback on those they do business with, and prospective traders can judge by from this whether they are trustworthy.
Crooks can, however, boost their own reputations artificially, to create a false reputation and can draw victims into fraudulent transactions. This typically involves creating bogus accounts and using these to leave positive feedback for the fraudster.
Advertisement
Software called NetProbe (Network Detection via Propagation of Beliefs) developed at Carnegie Mellon University, US, is designed to detect users who pump up their reputation in this way.
Auction site fraud, such as failure to deliver goods after a sale, accounted for almost two-thirds of 97,000 complaints passed on to law enforcement agencies last year by the US Internet Crime Complaint Center.
Systematic approach
鈥淭o the best of our knowledge, this is the first work that uses a systematic approach to analyse and detect electronic auction frauds,鈥� says Christos Faloutsos who created NetProbe with colleagues Duen Horng Chau, Samuel Wang and Shashank Pandit. The program scans publicly available logs of users鈥� previous transactions in search of fraudsters enhancing their own reputation.
The software plots a graph with lines representing the transactions between different user accounts. Transactions between accomplices and fraudsters create a pattern that sticks out like 鈥渁 guiding light鈥�, says Chau. NetProbe looks for a pattern known as a 鈥渂ipartite core鈥�. It occurs because auction site fraudsters typically conduct a disproportionate number of transactions with some users and none with others.
In a tests involving about a million previous transactions between about 66,000 eBay users, NetProbe correctly detected 10 known fraudsters, as well as more than a dozen other likely culprits. It can also produce its own trustworthiness ratings for users based on their previous transactions pattern.
Data mining
Similar data mining techniques are already used to detect credit card fraud or money laundering, notes Peter Sommer, a computer security expert at the London School of Economics, UK. 鈥淭he difference here is the researchers seem to be presenting this as an end-user tool as opposed to something deployed by specialist investigators,鈥� he told New 杏吧原创.
But Sommer says that implementing NetProbe could mean policy changes that some users object to: 鈥淚t would need a greater willingness to query sellers against whom no buyer had made a complaint.鈥�
Mike Joy, a computer scientist from Warwick University, UK, adds that fraudsters could potentially deceive NetProbe. 鈥淚t wouldn鈥檛 be difficult to change your behaviour to avoid creating that pattern,鈥� he says. 鈥淚f fraudsters networked amongst themselves it could obscure it.鈥�
Nonetheless, Faloutsos believes NetProbe ought to work in practise. 鈥淕etting around it will require more effort and cost, so fraud would be increasingly unprofitable,鈥� he says.
