Visual Semantics of Social Movements Online: ML/AI and Network Analysis for Social Media Messaging

Social movements have been using social media to share their narrative which supports their goals, values, and grievances, and it usually spreads across platforms in visual form. This research looks at how social movement images are structured into a narrative based on their visual semantic relationships. In the methodology, the images go through feature embedding, and zero‐shot captioning to create a network of semantic relationships. These relationships are represented in an ontological network where the labeled images are nodes and their descriptions are the edges semantically connecting them based on similarity scores. This network is analyzed using the Leiden community detection algorithm to understand the latent visual semantics supporting social movement images. Comparing the feature labels and visual semantic image captions graph to a graph of semantically similar images proved to be somewhat difficult for domain‐specific images like social movements. The analysis shows the various nuanced themes that arrive even within grouped classes.