It can be difficult to tell a substance abuser apart from the crowd despite the one in ten statistics of finding one, but analyzing social media activity can make this much easier and save billions of dollars as a result.

A test program called myPersonality was run on 150,000 participants’ Facebook accounts between 2007 and 2012, collecting data about each user’s interests, hobbies, and drug use levels, which included tobacco, alcohol, and recreational drugs. All participants were sorted into categories based on drug use frequency.

Once that experiment was complete, researchers recorded the status updates and likes data of millions of users and compared the results to the those of drug use. Reportedly, around 3,500 users fit into the likes and substance abuse combined category, while more than 1,000 users fit into the status updates and substance abuse combined category and less than 1,000 users fit into the triad category. With these fascinating results, researchers looked for recurring patterns in the overlapping groups, tested them, and achieved an 83.7% accuracy average across all drug types.

The experiment also identified several keywords used in social media posts that relate to the use of each drug. Increased use of words related to sex, swearing, and biological functions in posts upped the chances of all substance abuse types. Comparatively, labels used for space, females, and prepositions were linked to alcohol use, and anger and health words were linked to general hard drug use.

Interestingly enough, mention of hobbies and animation predicted a weak link between all types of drug use, just as male labels were not related to the use of hard drugs. With these telling results, users can more easily predict whether their friends and followers are abusing drugs or not, and can alert a trusted individual if evidence becomes more conclusive.