In JMIR public health and surveillance
BACKGROUND : Increases in electronic nicotine delivery system (ENDS) use among high school students from 2017 to 2019 appear to be associated with the increasing popularity of the ENDS device JUUL.
OBJECTIVE : We employed a content analysis approach in conjunction with natural language processing methods using Twitter data to understand salient themes regarding JUUL use on Twitter, sentiment towards JUUL, and underage JUUL use.
METHODS : Between July 2018 and August 2019, 11,556 unique tweets containing a JUUL-related keyword were collected. We manually annotated 4000 tweets for JUUL-related themes of use and sentiment. We used 3 machine learning algorithms to classify positive and negative JUUL sentiments as well as underage JUUL mentions.
RESULTS : Of the annotated tweets, 78.80% (3152/4000) contained a specific mention of JUUL. Only 1.43% (45/3152) of tweets mentioned using JUUL as a method of smoking cessation, and only 6.85% (216/3152) of tweets mentioned the potential health effects of JUUL use. Of the machine learning methods used, the random forest classifier was the best performing algorithm among all 3 classification tasks (ie, positive sentiment, negative sentiment, and underage JUUL mentions).
CONCLUSIONS : Our findings suggest that a vast majority of Twitter users are not using JUUL to aid in smoking cessation nor do they mention the potential health benefits or detriments of JUUL use. Using machine learning algorithms to identify tweets containing underage JUUL mentions can support the timely surveillance of JUUL habits and opinions, further assisting youth-targeted public health intervention strategies.
Benson Ryzen, Hu Mengke, Chen Annie T, Nag Subhadeep, Zhu Shu-Hong, Conway Mike
ENDS, JUUL, NLP, Twitter, e-cig, electronic cigarettes, electronic nicotine delivery system, infodemiology, infoveillance, machine learning, natural language processing, public health, smoking cessation, social media, tobacco, underage tobacco use