Drug Addiction: Spotting a veiled cry for help on social media

Cityscape with text bubbles and drugs spilling from container.Text: Treating addictions with online language processing

The expense of substance abuse in Canada is staggering. The latest comprehensive data from 2014 show the total yearly amount, including alcohol, tobacco, and opioid consumption, to be approximately $38 billion – or $1,100 per Canadian. These costs comprise healthcare, loss of productivity, and criminal justice. In 2014, alcohol and tobacco were by far the biggest culprits but the opioid crisis in recent years has brought narcotics like fentanyl into the limelight. Canadians are the second-highest per capita consumers of opioids in the world, which were the cause of approximately 4000 deaths in 2017 (an average of 11 per day); the federal government committed $100 million that same year to combat the crisis. While policy makers and healthcare experts scramble to find a solution, this scourge is most rapidly affecting young Canadians (15–24 years old) and Indigenous populations.

One of the key obstacles to tackling substance abuse within a community is the difficulty of offering tailored public health services to vulnerable populations. Traditional methods used by localities, such as health units, are essential but must evolve as issues continue to change. Vulnerable populations inhabit a variety of demographics including age, ethnicity, socio-economic status, and geographic location, to name a few, making it often difficult for healthcare providers to effectively reach those most in need of treatment.

Dan Lizotte and his research group in the Department of Computer Science at Western University have developed a machine learning model that is able to analyze discourse on popular social media sites to identify populations vulnerable to substance abuse. “We use natural language processing tools to see how people are communicating online – we can then help local healthcare providers, such as municipal health units, to figure out how best to tailor their services,” says Lizotte. Anonymous sites like Reddit give the Lizotte group a fairly accurate picture of the kinds of discourse used by people most likely to face addiction issues – health units can then align social media use with other current tools to more effectively reduce harm stemming from substance abuse.


Artificial intelligence techniques, such as the ones employed by the Lizotte group, are increasingly forming a part of the public health toolkit. The group’s complementary research into analyzing aggregated health-related data to assist with clinical decision making is further evidence of the power of advanced computing to simultaneously improve healthcare outcomes while reducing the strain on hospitals. Lizotte and his colleagues are now collaborating with health units around Ontario to roll out a front-end model that will be able to connect the health service providers with the vulnerable populations in their communities.

Listen to our related Podcast: Artificial Intelligence: Spotting a veiled cry for help on Social Media