Research & Innovation

Stevens Institute of Technology Researcher Unmasks the Power of Fake Online Content

Research team identifies ease of tricking rankings, outlines best defenses

Amazon, TripAdvisor, Yelp: these and other consumer services depend upon accurate, crowd-sourced content to guide users, build trust and grow their user bases.

But what if those posted reviews aren't all truthful? Can fake, planted positive reviews — or negative reviews fraudulently posted in competitors' entries — change rankings and influence consumers' choices?

"Yes, unfortunately, " says Stevens Institute of Technology business professor Ted Lappas, who has just completed a sweeping survey of online reviews in new research published in the journal Information Systems Research.

"It is surprisingly easy to intentionally push a business up in review-site rankings using fake reviews."

The right number to tip the scales

Researchers estimate 15 to 30 percent of all online reviews may be falsified and planted for positive or negative purposes.

Lappas, who has been following research in fake online review content for a decade, set out to discover whether these efforts make any difference — and, if so, how to combat them.

"We are not talking about simply a semantic discussion any longer," he points out. "Studies have proven this effect goes far beyond that: it is now economic fraud, making a difference in the life and death of real businesses."

To analyze the phenomenon, Lappas, Stevens graduate student Georgios Valkanas and associate faculty member Gaurav Sabnis studied roughly 2.3 million online reviews of nearly 5,000 hotels in 17 U.S. cities.

Visibility is key, the team found. More and more positive reviews obviously push certain hotels or restaurants higher in search rankings, but Lappas was surprised by how little effort is required to do so.

"Just 50 or so additional positive reviews planted on a site are sometimes enough, we found, to tip the scales and drive a particular property to the number-one ranking in their market," he notes.

And it takes even fewer fake reviews — perhaps as few as ten or 20 — to push a business into, say, the top five in its category if that business is also simultaneously spamming some of its top competitors with fake negative reviews.

"This is concerning, and we call for these sites to be aggressive about fraud detection," says Lappas. "You have to go beyond analyzing the review text. You have to go beyond looking for unusual bursts in the number of reviews of a single business. You have to look at visibility, at the rankings."

One deception strategy works best

The most effective fraud strategy of all, the Stevens team found, was a mixed strategy — a business planting fake positive content and also seeding higher-ranked competitors' entries with made-up negative reviews.

"It's very inexpensive these days to simply pay a service or writer to create these fake reviews and flood the sites with them," he points out. "And there's usually little penalty for doing so. All you have to do is post fake negative reviews for strategically selected competitors, while injecting positive reviews for your own business."

"Using this strategy, you can make a big difference with a surprisingly small number of reviews.”

Even verified-reviewer programs such as the one offered by Amazon don't blunt this effect, Lappas found.

"Verified reviewers often receive discounts or free products to review, and this can erode users' trust in what they're writing. There are also a host of privacy issues that come along with creating a verified-reviewer system," he notes.

The best defense

In their research paper, the Stevens team suggests effective fraud-detection strategies for operators of review websites, including avoiding overemphasis on very recent reviews — a common practice.

"There are several ways operators can attack the problem," Lappas explains. "One is to create a 'buffer zone' of weeks or months before brand new reviews appear online, so that algorithms and human monitors have the time to determine whether they are likely truthful or not. Our simulations provide strong evidence that this would be effective."

Fraudsters will be deterred from posting false reviews, he adds, if there is a delay before reviews are posted live to review websites.

"For those who decide to go ahead with fraud anyway, this buffer zone then makes it very difficult to strategically choose competitors to target. An attack you design today might be far less effective only weeks later, when the buffer empties and the review landscape is completely different than what it is today," Lappas says.

Another tactic is to closely monitor the visibility of businesses over time.

"If a small business suddenly shoots up the charts quickly in rankings, it should be a red flag, even if the number of reviews for this business has not changed significantly. The change may be due to reviews injected to competitors." he says. "Software we are developing can catch this, and flag influential reviews for closer inspection."

Proprietors can also fight back

Finally, Lappas also suggests business owners take matters into their own hands by responding when possible to false or potentially false negative reviews. Interestingly, while TripAdvisor, Yelp and other platforms do allow business owners to respond to negative reviews, Lappas' research finds that roughly two-thirds of negative reviews receive no response at all, and of the remaining one-third that do, only 5 percent of the responses challenge the user review.

"Many of the current owner responses seem to be cut-and-paste jobs, simple apologies or thank-you's for raising a concern," he says. "They're not really addressing the content of the real or potentially fraudulent complaint."

On the other hand, Lappas cautions that replies mustn't cross the line into over-aggressiveness, either.

"There is an optimal way to respond to a customer complaint, and you sometimes see these business owners stepping far across that line of courtesy," he chuckles. "An over-aggressive response that protests too much tells you something else about the property: that maybe the user's complaint was real, after all."