Consumers today are heavily dependent on online reviews to make informed choices about what to buy. In fact, studies show that as many as 90 percent of consumers read online reviews before making financial decisions, and nearly 70 percent trust these opinions.
Given their importance, how do you tell if the reviews are from genuine customers?
Subodha Kumar, director of the Center for Data Analytics and professor of Marketing and Supply Chain Management at the Fox School, developed an approach to detect fake reviewers on online digital platforms. In his paper published in the Journal of Management Information Systems, Kumar proposes an algorithm that analyzes the behavior of reviewers on a set of key features to help differentiate between the real and the fake.
“A user who reads a negative review of a restaurant is likely to trust the message, even though it was written by a stranger,” Kumar says. “One convincing review can often persuade consumers to shift their brand loyalty or drive several extra miles to try a new sandwich shop.”
This gives firms a strong incentive to influence their online review ratings. “Business owners inject their public ratings with a positive bias,” says Kumar. “They use fake accounts or paid reviewers to either promote their offering or strategically denounce competitors’ products.”
In studying a dataset from Yelp, a popular restaurant review platform, Kumar observed a striking difference in the way spammers interact on online platforms. “Even though individual reviews by a spammer may look genuine, collectively we can capture anomalies in the review patterns,” Kumar says, “In fact, they are remarkably skewed.”
By analyzing this pattern of behaviors, Kumar’s approach to detecting review manipulation can not only improve the experience of consumers across industries but also increase the credibility of reviewing platforms like Yelp.
Kumar considers six distinct features of every review in the data set:
- Review gap: Spammers are usually not longtime members of a site, unlike genuine reviewers who use their accounts from time to time to post reviews. Thus, if reviews are posted over a relatively long timeframe, it suggests normal activity. But when all reviews are posted within a short burst, it indicates suspicious behavior.
- Review count: Paid users generally generate more reviews than unpaid users. In other cases to avoid being detected or blacklisted, a spammer could post very few reviews from one account and create a new account.
- Rating entropy: Spammers mostly post extreme reviews since their goal is either to artificially improve a particular company’s rating or to bring a bad reputation to its competitors. This results in high entropy—or drastic randomness—in fake users’ ratings.
- Rating deviation: Spammers are likely to deviate from the general rating consensus. If genuine users fairly outnumber spammers, it is easy to detect instances where a user’s rating deviates greatly from the average ratings from other users.
- Timing of review: One strategy spammers may use is to post extremely early after a restaurant’s opening in order to maximize the impact of their review. Early reviews can greatly impact a consumers’ sentiment on a product and, in turn, impact sales.
- User tenure: Fake reviewers tend to have short-lived accounts characterized by a relatively large number of reviews and handles, usernames or aliases designed to avoid detection.
After considering these variables individually, the algorithm then looks into the way the variables interact with each other. It employs techniques like supervised machine learning and accounts for the overall review behavior of a user to provide a robust and accurate analysis.
Kumar’s methodology can also be deployed to post the information of the spammers in real-time. Digital platforms like Yelp could develop a spam score using these key features for each reviewer and share it with business owners and consumers, who can subsequently be tagged or filtered.
“The issue of opinion spamming in online reviews is not going away and detecting the perpetrators is not easy,” says Kumar. But developments in approaches like these, he says, “offer great insights to businesses, allowing them to create more effective marketing strategies based on the sheer volume of genuine, user-contributed consumer reviews.”