Idiap’s Social Computing Group received a paper award at Pervasive Health 2019. Their research focused on drinking practices pictured on Instagram and classified thanks to hashtags, images, and artificial intelligence.
On social media, people share daily activities, including alcohol consumption, through images and text. Using this data to extract relevant information can help to understand alcohol consumption patterns, especially those indicating negative drinking behaviour. Such tools are particularly helpful for both researchers and health policymakers. Thanh-Trung Phan, Skanda Muralidhar, and Daniel Gatica-Perez from the Social Computing Group decided to use Instagram images and hashtags as a resource for their research. They used machine-extracted textual and visual cues to reveal and analyse drinking trends. This research received the Honorable Mention Paper Award at the International Conference on Pervasive Computing Technologies for Healthcare (Pervasive Health) held in Trento, Italy.
What can we learn from hashtags and images?
This work used a five-year dataset from Instagram. Thanks to computer algorithms, researchers were able to reveal trends linked with casual drinking and possible negative drinking. To distinguish these concepts, they used two different concepts evoked by hashtags: #drink and #drunk, and gathered information related to each one of them. Their analysis revealed 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. A further analysis showed that #drunk posts have a higher chance of being perceived as potentially objectionable. A random forest classifier also showed that #drink and #drunk posts can be discriminated with accuracy up to 82.3%, confirming the relevance of such automated tools to extract information.
This work was conducted in the framework of the SNSF-funded Dusk2Dawn project, which studies nightlife as experienced by young people, integrating research in social computing, public health, and human geography.
Texte by Ideap