Cornell Computer Students Create Program to Keep TripAdvisor Honest

A group of Cornell University computer science students have created a program that can determine with nearly 90% accuracy whether hotel reviews on travel sites like TripAdvisor or Expedia are real or promotional.

The findings, presented at the 49th annual Association for Computational Linguistics on Sunday, showed that fake reviews can be picked out because they follow certain patterns.

According to Cornell's news service, researchers found that real reviews tend to coalesce around concrete words like "bathroom" and "price" while fake reviewers tend to spend more time setting the scene with words like "vacation" and "my husband." Essentially, fake reviewers make a big effort to seem like real people. Real people don't.

TripAdvisor boasts that it has programs designed to pick out fake reviews, but many still slip through. Picking them out by hand is also nearly impossible: The Cornell researchers found that humans identify fake reviews no better than chance.

The new software is being pitched as a potential "first-round filter" to get rid of opinion spam. TripAdvisor, at least, could use the help.

A cursory search of reviews of hotels in Mexico City finds plenty of entries that rather neatly match the fake review criteria.

A review of the rather obscure Suites Contempo hotel - written by a first time contributor to the site leads with this: "The great thing about this boutique hotel is service! Service you will not get at a large chain hotel where you are just another guest. Over the last 4 years, I have easily spent 3 months in this hotel, so I consider it a second home."

Is the review fake? It certainly hits the marks, but I'm not the one to ask (full disclosure: I'm human). Someone better go get the Veritasinator 3000.