My local newspaper just featured a story about Yelp reviews of local restaurants. In it, the author claims that the online environment is “rife with false reviews.” And a little over a month ago, my cousin-in-law sent me a link to yet another news story about fake reviews. Of course, fake reviews remain a problem — as the popular media continually reminds us. Given the anonymity that is afforded by posting on the internet, the PBS story’s correspondent describes the quest to identify fraudulent reviews as an “arms race” that will never be won.
However, as I was searching for recent academic work on this topic, I came across a paper that offers a slightly different perspective. In this study, researchers asked 80 people to write fake reviews. It was kind of like a homework assignment. They gave each participant the website of a particular hotel and told them to write either a positive, negative, or neutral review about their hotel for a site like TripAdvisor. After that, the researchers asked their participants to talk about how they went about composing those fake reviews. Interestingly, their participants used a variety of textual borrowing strategies, ranging from copying + pasting bits of existing TripAdvisor reviews for the same property, to paraphrasing from a model review. The findings of this study suggest that fake reviews actually *share many textual features with real reviews,* since they may be very closely derived from them. Of course, this is just one of many reasons that makes some of them so difficult to detect.