The slogan "If you drink, do not drive" has become popular among users of social networks with their own variant: "If you drink, do not tuitees". Write messages with a few drinks above is not the best option. Now a group of researchers from the University of Rochester (USA) has taken advantage of this to create an algorithm capable of detecting the tweets that are written under the influence of alcohol. Your system also allows you to locate the place from which the tweets were sent. The aim of the study is to predict social patterns and habits related to alcohol consumption to avoid public health problems, reports El País of Spain.
But how can an algorithm to detect this type of tweets? The method is simple. First, messages sent for six months in a given area (the researchers chose the area from New York to Monroe County, USA) are collected. During that time they leaked all tweets in which the word alcohol is mentioned or related, such as drinking, drunk, beer or party. To analyze in detail, they hired workers who decided if the message referred to be drinking alcohol and if it was just in the moment in which he tweeted.
The result found 11,000 tweets sent under the influence of alcohol. With these data he trained the algorithm so that from them, were able to find all messages that users send while they drink.
"This method has great potential," explains one of the authors of the study and researcher at the University of Rochester, Ann Marie White. "In the future, for example, it might be possible to notify you if a friend is tweeting under the influence of alcohol, so you can care for and reduce risks. In the same way you would in the real world if you were hanging out with that person. "
The study goes a step further and also seeks to know whether the tweeters are drinking at home or outside. The method is similar to the first but analyzing messages containing phrases like "Finally home!" And words like bath, sofa or television. Then the process is refined with the participation of workers, who decide if it appears that the tweet was written or not from home. Across the two types of data, the algorithm is able to know when and where people who are tweeting drinks.
The long-term goal of these researchers is to be able to protect the young people of the risks involved in social networks publish under the influence of alcohol. "For example, you can lose a job by posting a photo of yourself drinking while wearing the logo of a company," she argues.
White and his team have big plans for this method. One of them is to create an application based on these algorithms, could warn users when someone is talking about while drinking alcohol or tweeting. "Although we should investigate whether it is feasible and if people accept it."