Abstract: Mining Twitter to Assess the Public Perception of Controlled Substances (Society for Prevention Research 24th Annual Meeting)

608 Mining Twitter to Assess the Public Perception of Controlled Substances

Friday, June 3, 2016
Regency B (Hyatt Regency San Francisco)
* noted as presenting author
Chris Delcher, PhD, Adjunct Assistant Professor, University of Florida, Gainesville, FL
Jian Bian, PhD, Assistant Professor, University of Florida, Gainesville, FL
Introduction: More people prefer social media to other sources for obtaining information in real time. As of January 2014, 74% of online adults use social networking sites. User generated content on social media such as Twitter is a valuable resource because it can provide a source for gleaning information about people’s daily life to answer scientific questions. These new data sources can expand the range of what can easily be measured and provide new types of information for mining useful knowledge.

We present a framework of social media analysis to examine the public’s perception of controlled substances. We compare public’s view on opioids, specifically oxycodone, and heroin, to assess population-level factors associated with the epidemic of opioid abuse.

Methods: We collected 863 million random samples of all public tweets posted in 2014 (from February to November) and 2015 (from January to August). Of which, 280 million are written in English. These 280 million tweets represent the analytic data set. First, we used keywords (i.e., “heroin”, “opioid”, and “oxycodone”) and hashtags (i.e., “#heroin”, “#opioid”, and “#oxycodone”) to identify tweets relevant to the discussions. Second, we performed trend analysis, including comparison to Google trends, to understand when and how these gained wide interests. Third, we conducted sentiment analysis to assess the public’s response and applied the Latent Dirichlet allocation (LDA) approach to find latent topics associated with these discussions.  

Results: Out of the 280 million tweets, we found 10,964 tweets are relevant to the discussions of heroin (n=10,024) and opioid (n=940). The trends of these discussions in Twitter are highly correlated with their Google trends. This is expected as users express their opinions regardless of which online resource they use. Our sentiment analysis shows that users have mixed feelings (positive vs. negative emotions) towards opioid use. Further, through topic modeling, we find that the general public recognizes and is concerned about pain and overdose.

Conclusions: We are surrounded by social media, and information generated through such sources can be used to understand public opinions on many health and health policy issues. This is important as public opinion is a component of public policies. When the public believes a problem is a pressing concern, policy action is more likely to occur and to succeed.  For example, legal policies on the distribution of Naloxone, a drug known to reverse an opioid overdose, are changing in response to the epidemic.