140 Characters of Intoxication: Exploring the Prevalence of Alcohol-Related Tweets and Predicting Their Virality

Abstract

Past research has suggested an associative relationship between social media use and alcohol consumption, especially among the younger generations. The current study takes a generalizable approach to examining the prevalence of posting about alcohol on a popular social media platform, Twitter, as well as examining the predictors of a tweet’s virality. We content-analyzed more than 47.5 million tweets that were posted in March 2015 to explore the prevalence of alcohol-related references, and how alcohol-related references, tweet features (e.g., inclusion of hashtags, pictures, etc.), and user characteristics (e.g., number of followers) contribute to the tweet’s virality. Our findings showed that during March 2015, about two of every 100 tweets in the United States were alcohol-related; whereas the majority of those referenced intoxication. In addition to tweet features and user characteristics, the prevalence of alcohol-related references in a tweet positively predicted the number of likes it received, yet negatively predicted the number of retweets. Given prior evidence supporting the association between social media use and alcohol consumption, the prevalence of alcohol references in tweets and how that contributes to their virality offers insights into the widespread phenomenon of glorifying alcohol use and excessive drinking via social media, pointing to potential negative health consequences.

“Can't Wait to Blackout Tonight”: An Analysis of the Motives to Drink to Blackout Expressed on Twitter

“…In addition to severe long‐term negative consequences (i.e., cancer; Connor, ), alcohol use is also associated with a range of short‐term negative consequences including alcohol‐induced memory loss (Hingson et al., ; Valenstein‐Mah et al., ). Social networking websites have become increasingly popular and are a common outlet for individuals to discuss alcohol and substance use (Alhabash et al., ; Boyle et al., ). But to date, no studies have examined the discussion of alcohol‐related memory loss (i.e., blackouts) on Twitter, despite the value such exploration may have for our understanding of blackout‐related attitudes and behaviors.…”

“…For example, a study of 11,236 university students found that 97% used social networking websites and that 71% of these students reported posting alcohol‐related content (Erevik et al., ). On Twitter specifically, Alhabash and colleagues () found that 2% of the nearly 50 million Tweets published in the United States in March 2015 included alcohol‐related content (Alhabash et al., ). Exposure to and engagement with alcohol‐related social networking content may be linked to alcohol use.…”

“…Although drinking‐related posting is common on Twitter (Alhabash et al., ), there is a relative paucity of literature examining the content of Tweets. To date, Twitter content research has often either focused on temporal variations in Tweets (i.e., differences in the number of Tweets during certain time periods) or the sentiment of the Tweets (i.e., positive or negative content).…”

“…Rather than testing a priori hypotheses or seeking confirmation of specific theoretical frameworks related to intentions and motives for drinking broadly, due to a specific focus on blackouts, our goal was to generate hypotheses that could be more rigorously tested in future studies with superior sampling strategies and statistical tests. Twitter is an ideal platform to collect these preliminary data, as an estimated 2% of Tweets contain alcohol‐related content (Alhabash et al., ), Twitter posts are common among the highest risk age‐group (young adults), and the Tweets are written unprompted by researchers. Furthermore, researchers have noted the limitations of retrospective assessments of motives for drinking behavior (Gmel et al., ) and using Twitter data allows for an ecologically valid examination of motivations for a behavior prior to the actual occurrence of that behavior.…”

A Machine Learning Approach to Predicting Alcohol Consumption in Adolescents From Historical Text Messaging Data

“…As such, we apply doc2vec to samples of text messaging from individual participants in the BlackBerry Project study, then leverage the resultant vectors as input to a standard logistic regression algorithm to predict whether the individual self-reported alcohol use during the time period over which the texts were collected. We achieve above 76.5% accuracy on this binary classification task without the need for curating domain-specific keyword lists or manually filtering individual samples, as are typical in studies of this nature [14,15,16].…”