Researchers at Cornell have developed software that can detect phony online reviews. It could be used to combat the booming business of deceptively using websites like TripAdvisor and Yelp to promote a businesses or badmouth its competition.
The software finds what they call "Deceptive Opinion Spam" with up to 90% accuracy by finding words more likely to be used in fake writeups. Humans, by comparison, could detect the fakes just as easily by wearing a blindfold and choosing at random.
The researchers developed the algorithm by looking at the types of words used in both fake and real hotel reviews. Turns out the fakes use words that develop a sense of place — "family," "husband," "vacation" — while the real ones are more specific about hotel features. (Here's the report if you want to see all the metrics they used to determine the word sets; warning, PDF.)
So how could this be used? To help identify reviews produced by professional opinion spam farms.
Will this tool be able to stop professionally-created fake online reviews? Les Inrocks has a whole story (French only, see Google Translate) about how they got quotes from multiple companies that traffic in the fake online review business.
One of the companies based out of Madagascar uses an "industrial" approach and offered one of its full-time 75 employees for 550 euros a month. That single person is capable of dropping 25,000 fake reviews in three months.
Another agency, based of out Paris, sends employees out to coffee shops (for the variable IP addresses). They gave a quote of 3750 euros (before tax) for 250 comments. The company showed Les Inrocks some fake reviews they had written on Tripadvisor, one of the largest hotel review websites in the world, and explained their techniques: "Sometimes the hotel is described as 'fantastic' or 'the epitome of hospitality,' sometimes reviews are negative as to not arouse the suspicions of the webmaster." For example, one of the reviews of a hotel reads: "Sometimes I wish I had more outlets in the rooms. But it's true that I took all my electrical appliances LOL."
Said the representative, "You can quickly get caught, you have to go very quietly... Do not put a lot especially at the start. Competitors identify it right away. If it was a hostel, we will leave comments that appear young, with pretty spelling mistakes."
· Software Picks Out Fake Online Reviews [GizMag]
· Comment des entreprises françaises fabriquent et vendent de faux avis sur le Net [Les Inrocks via @clotildenet]
[Photo: fstorr / Flickr]