#Drink Or #Drunk

Abstract

The understanding of alcohol consumption patterns, especially those indicating negative drinking behavior, is an important issue to researchers and health policymakers. On social media, people share daily activities, including alcohol consumption, representing these moments through images and text. This work, using a five-year dataset from Instagram, analyzes what machine-extracted textual and visual cues reveal about trends of casual drinking (concepts gathered around #drink) and possible negative drinking (concepts gathered around #drunk). Our analysis reveals that #drunk posts occur more frequently in party occasions and nightlife locations, with a higher presence of people, while #drink posts occur at food locations, with a higher presence of drink containers. Manual coding further shows that #drunk posts have a higher chance of being perceived as potentially objectionable. A random forest classifier shows that #drink and #drunk posts can be discriminated with accuracy up to 82.3%. These results have important implications for alcohol research among youth. CCS CONCEPTS • Human-centered computing → Ubiquitous and mobile computing; Ubiquitous and mobile computing design and evaluation methods;