The epidemic of substance use (SU) and substance use disorder (SUD) in the United States has been evolving for decades. Both prescription and illicit drugs have been involved in overdose deaths over the years, with notable increases in synthetic opioids (eg., fentanyl & analogs) and psychostimulants (eg., methamphetamine) in recent years. The emergence of high-potency novel psychoactive substances (NPSs), such as fentanyl analogs, have drastically contributed to rising deaths, and adversely impacted treatment engagement and response.
A key element to tackling the crisis is improved surveillance. Specifically, there is a need for establishing novel approaches to provide timely insights about the trends, distributions, and trajectories of the SUD epidemic, as traditional surveillance approaches involve considerable lags. Many recent studies have identified social media (SM) as useful resources for conducting SU/SUD surveillance. Many people use SM to discuss personal experiences, provide advice, or seek answers to questions regarding SU/SUD, resulting in the generation of an abundance of information. Such information can be characterized, aggregated and analyzed to obtain population- or subpopulation-level insights, at low cost and in near real time. However, converting SM data into timely, actionable knowledge is non-trivial since the data is big, complex, and noisy, requiring the development of advanced, automated artificial intelligence methods.
This presentation will highlight ongoing and past work on developing NLP and machine learning methods for effectively leveraging social media data for substance use research.
Presenters
