It’s gotten to the point that when looking at some vacation rentals, if it only has a few reviews, some people ignore five-star reviews entirely and focus on one-star reviews (to check for real complaints versus griping or malice) and three-star reviews (usually a good guide to the pros and cons of something).
There’s good reason to be skeptical – as the New York Times reports, more than a few companies have taken to hiring people to write rave reviews of their products or services.
Given that cheaters abound, it is in the interest of companies such as TripAdvisor and HomeAway to make sure that reviews are genuine. If they’re not, and customers learn they can’t trust reviews, they’re likely to take their business elsewhere.
And this kind of “opinion spam” is getting more advanced – meaning that it’s harder for people to tell if a review is a real one, or just a fake rave for hire.
Enthusiastic endorsements don’t always come in prose-form. There are a number of shops that have popped up to sell Facebook fans, Twitter followers, and even +1 buttons for the recently arrived Google Plus.
This is why the HomeAway’s and FlipKey’s of the world should be interested in research from a team at the Computer Science Department at Cornell, which focuses on algorithms designed to spot “deceptive opinion spam” – fake reviews that have been written to sound authentic.
Their research, which was recently presented at the 49th Annual Meeting of the Association for Computational Linguistics focuses on several different techniques for spam-spotting, as well as combinations of the same. Using these techniques, they were able to spot fake reviews with about90% accuracy – far better than the human control group, which did no better than chance.
Note that this research was directed at finding false positive reviews. The group’s next step is to refine their techniques and then extend them to include false negative reviews, as well.
If their research is successful, it may well be implemented in large sites to improve regulation of reviews and get the spam out.