Artificial intelligence has calculated for alcoholics social media posts
American researchers have created artificial intelligence, which by the posts and likes of Facebook users to determines if they suffer from tobacco, alcohol or drug addiction. The maximum accuracy of the program reaches 86 percent. The work of scientists published in the preprints server ArXiv.org.
According to statistics, one in ten Americans aged 12 years and older suffers from some form of addiction (substance use disorder, SUD). In Russia, about 8.5 million people are abusing drugs and psychotropic drugs, and more than 20 millions of Russians are dependent on alcohol.
But researchers recently found a correlation between personality traits and the propensity to consumption of different substances. So, regular tobacco Smoking people are much more open to the experience, but less conscientious than non-smokers (here honesty means self-discipline, commitment and desire to achieve goals). Alcohol consumption, in turn, is positively correlated with sociality and agreeableness.
As people using social networking, according to the Internet a lot of information about their interests and personality traits, the authors of the new work suggested that the posts and likes can also point to bad habits of users.
The researchers applied a machine learning algorithm, trained using the three databases that were collected in the period from 2007 to 2012, the application of psychological tests myPersonality. The first contained 21 million records 100 thousand users of Facebook, the second — 5 million likes 250 million users; the third kept the information about the dependencies have 13.5 thousand users. For learning, these datasets were combined in different ways.
After training, the system learned to recognize the presence of harmful habits in a person. The likelihood of Smoking tobacco is determined with a maximum accuracy of 86 percent, the probability of drug use was 84 per cent, the probability of alcohol consumption was 81 per cent.
While scientists have found a correlation between the content of the posts, the interests of users and different kinds of dependencies.
For example, the algorithm has calculated that heavy drinkers and cigarettes often use words connected with movement “the car” or “go”. Words relating to anger (“hate”, “kill”) and health (“hospital”, “tablets”), is positively associated with drug use. In addition, it was found that drinking alcohol people love the movie “V for vendetta”, and drug addicts listen to Radiohead, The Cure and Depeche Mode.
However, it should be noted that a positive correlation in this case does not indicate a causal relationship. Moreover, the overlapping data set was much smaller than the volume of each of the three databases separately — it contains information only about 3508 users. In order to improve the accuracy of the results, we need much more information.
Recently, scientists have established that others do not see large differences between the behavior of a sober and drunk man, while the one who was drinking alcohol, observes in himself a radical change. Also, the researchers were able to find the gene associated with craving for alcohol.