Through the Eyes of Instagram: Analyzing Image Content utilizing Meta's Automatic Alt-Text

Multimedia communication has become an essential part of social media, with images representing a significant part of the content on most platforms. This study investigates image content on Instagram through Meta’s internal image classification algorithm, Automatic Alt-Text (AAT). Our approach differs from research on data from comments and hashtags because of the use of actual visual descriptions as the means of understanding the kinds of the content published on the network. Our analysis of 200k posts reveals 1,471 unique tags being used to characterize image content on Instagram, representing mostly objects, food, animals, locations and other common components of social media photos. Notably, we found that content about personal aesthetics is highly popular on the platform, with person and selfie being respectively some of the top two most common tag and post categories, being also highly related to other tags such as makeup, lipstick and eyeliner. Furthermore, we explored the connections between tags, representing very popular content trends within the network. Finally, we uncover substantial differences in posting behavior of influencers and news pages when compared to regular users, observing they post more frequently and about more specific content, suggesting what may attract more engagement on Instagram.