Deep learning could help catch drug dealers who use social media to connect with customers, according to a study conducted at UC-San Diego and published June 14 in the Journal of Medical Internet Research.
The authors focused on Instagram. Commenting on their findings, they point out that its platform continues to allow dealers to post despite increased heat from regulators and policymakers.
For the research, Timothy Mackey, PhD, and colleagues used a Web scraper to identify 1,228 posts from 267 drug dealers, then compared their deep-learning model against three machine-learning techniques (random forest, decision tree and support vector machine).
Deep learning performed the best, achieving 95% accuracy at correctly identifying illicit sales activity involving oxycodone, methamphetamine, LSD, Xanax and others.
Manually annotating all analyzed posts to characterize relevant conversations, the researchers further found that it’s not uncommon for sellers to invite buyers to complete transactions by switching from social media to direct communications.
“These initial results are alarming and generally conform to existing nonscientific investigational news reports on the subject, in addition to published research on illegal drug sales on this and other social media platforms such as Twitter,” Mackey et al. write in their discussion section.
They reference congressional testimony given in 2018 by Mark Zuckerberg. The Facebook chairman and CEO said his company was developing AI tools to proactively target social-media drug dealing.
Subsequently, the authors note, Instagram as well as Facebook said they’d suspended accounts and blocked opioid-related hashtag searches.
“However, our study indicates that drug sellers continue to populate Instagram despite these actions and that these communities have changed their use of hashtags possibly to avoid detection,” they write. “Hence, to carry out the legislative intent … to promote patient safety and prevent substance abuse behavior, there is clear need for innovative technology solutions that have high accuracy and are scalable and can help all parties (including technology companies, regulators, and law enforcement) detect, classify and take action against digital drug dealers.”
The study is available in full for free.