We all know that customer reviews can be prone to, shall we say, a little positive engineering. What if you could gather genuine opinions about a restaurant, or product before you commit your cash? Well, a new system developed at the University of Rochester might be able to offer just that. The "nEmesis" engine uses machine learning, and starts to listen when a user tweets from a geotagged location that matches a restaurant. It then follows the user's tweets for 72 hours, and captures any information about them feeling ill. While the system isn't able to determine that any resulting affliction is directly connected to their restaurant visit, results over a four-month period (a total of 3.8-million analysed tweets) in New York City found 480 reports of food poisoning. It's claimed these data match "fairly well" with that gathered by the local health department. The system's creators admit it's not the whole picture, but could be used alongside other datasets to spot potential problems more quickly. The only question is how long before we see "sabotage" tweets?