READ MORE / " title="'AFTER SOME TIME, I can estimate whether I can trust this rater or not,' explains Yan Sun, an assistant professor of computer engineering at URI and who recently was awarded a five-year National Science Foundation research grant. READ MORE /"/>A team of computer engineers at the University of Rhode Island has requested patent protection of a technology to combat the manipulation of user-generated ratings of merchants and products offered on eBay, Amazon and other Web sites.
Qing Yang, a professor of computer engineering; Yan Sun, an assistant professor; and Yafei Yang, a graduate student, last month filed a provisional patent for their technology.
Amazon, eBay, NetFlix, Yahoo Shopping, CitySearch and many other virtual marketplaces allow users to rate and review the books, movies, restaurants and other products available on the Web sites, and in some cases to rate the merchants selling them.
These ranking systems have a significant commercial impact: NetFlix customers are more likely to rent an obscure independent movie with a five-star rating from other users, while CitySearch visitors are more likely to avoid a restaurant given bad reviews by prior users.
But as ratings have risen in popularity and commercial importance, scam artists, spammers and others have increasingly found ways to manipulate the systems. In some cases, those perpetrating the fraud are companies or individual merchants looking to boost sales or increase traffic to their Web sites. In other instances, user-generated rankings are manipulated by criminals looking for an easy score.
Scam artists on eBay typically boost their positive feedback rating by selling hundreds of low-cost items, and then use that sterling reputation to dupe unsuspecting consumers by pocketing the money on a few big-ticket sales without ever mailing the items. Book publishers will sometimes boost ratings of a new book by paying people to give it glowing reviews in online discussion forums.
“The problem is if I call all of my friends and say, ‘I want to boost the rating of this product,’ ” said URI’s Sun. “I may recruit 30 people and then ask them to all give it high ratings, and then that product’s ratings will be much higher than what it should be. Or I can call my friends and say, ‘I hate this product – lower its ratings.’ So it’s very easy to do.”
Administrators of e-commerce Web sites are constantly seeking better methods to fight the manipulation of ratings and reviews. Auctioneer eBay trolls its Web site with sophisticated fraud-detection algorithm tools to spot suspicious activities and individuals who may be attempting to inflate their feedback. But fraud still occasionally occurs on eBay, and other Web sites have fewer controls.
The manipulation-busting technology patented by the URI team works in two steps, Sun said. First, it employs an algorithm that flags voting or review patterns that suggest fraud. A flurry of multiple users rating a single product at once, overly similar voting by separate users, and users who rate or review without spending time viewing a product all send a distinct signal amid the background noise of fair ratings, Sun said.
As a second step in the process, the program also tracks users of e-commerce sites who have been caught manipulating ratings in the past and gives their input less weight when averaging out future ratings, Sun said.
“After some time, I can estimate whether I can trust this rater or not,” she explained. “You have ratings, for example, from 50 people, and each of them will have a trust score. When you combine their ratings, it’s not a simple average – it’s a calculation based on the rating they give and the amount of trust I have in that rating. So the raters have low trust value if they get signaled a few times, and they have very little influence on the final rating.”
The URI team will formally introduce the technology at the First International Workshop on Trust and Reputation Management in Massively Distributed Computing Systems – an industry conference that will be held in June in Toronto.
The computer engineers hope to file a formal patent within a year, Sun said, with the hope of bringing the technology to market soon afterward.
In the meantime, they are preparing to run a challenge in which computer engineering teams from all over the world will be asked to download and manipulate ratings to the best of their ability. The team that can manipulate them the most effectively will win a prize.
Sun said the purpose of the exercise is to test the URI team’s algorithm against many forms of high-level manipulation, and to collect data on different attacking methods and use them as benchmarks to help other researchers in the field to test their own algorithms.
