Text Analytics Can Save Us From Fake Reviews

  1 m, 47 s

With the rise of service-rating websites like TripAdvisor and Yelp, along with online marketplaces like Amazon and Newegg, has come a new grey-market industry: fake review writing. Thousands of freelance writers across the globe sell their services each day writing positive (or negative) reviews of goods and services they have never experienced. 

A 2013 investigation found that “reputation shops” were all too eager to offer their services using techniques ranging from bribing customers with $50 gift cards to employing people in Bangladesh, the Philippines, and Eastern Europe to produce “buckets of praise for places they had never seen in countries where they had never been.”

Fake reviews are also used to attack other companies. A leading British historian, for instance, had to pay damages after a libel suit against him alleged that, under a false name, he posted defamatory reviews of rival books and positive reviews of his own.

The news is not all bad: a quick online search comes up with a host of advice columns offering suggestions on how to spot fake reviews, including tips such as “Consider the length and tone of the review”, watch for someone “using an abundance of personal pronouns”, and “being just plain over-the-top”. 

But detecting fake reviews on your own is at best time-consuming and unreliable; according to researchers at Cornell University, the average person can only detect fake reviews 50 percent of the time.

Using text analytics and machine learning techniques to detect language people use when being deceptive, the same group of researchers developed an algorithm capable of detecting whether or not a hotel review is real or fake with a 90% accuracy rate. (In case you’re wondering, deceptive writers tended to use more verbs and focused on family and activities, while real reviewers used more punctuation and focused more on the hotels themselves.)

In an age where online reviews are one of the most influential sources for travel planning, being able to weed out people trying to game the system is crucial for sites like TripAdvisor. Researchers at Cornell hope to branch out and further develop their software in order to detect fake reviews in other fields.

Categories: Insights, Text Analytics