Congratulations to CCPP faculty affiliate Dr. Ayşe D. Lokmanoğlu, whose new co-authored article, “Topic Modeling of Video and Image Data: A Visual Semantic Unsupervised Approach,” has just been published in Communication Methods and Measures.
The paper introduces VisTopics, an innovative computational framework for analyzing large-scale visual datasets by combining AI-powered image captioning, unsupervised clustering, and topic modeling. Applying the method to over 450 NBC News videos and 7,000 socially mediated news images, the study demonstrates how AI-driven captioning and Latent Dirichlet Allocation (LDA) can reveal the hidden semantic structure of visual narratives—from political unrest and environmental disasters to consumer culture and social life.
Lokmanoğlu and co-author Dror Walter (Georgia State University) show that VisTopics enables interpretable, scalable, and human-readable analysis of images and videos, bridging the gap between visual representation and semantic meaning. The framework provides powerful tools for mapping and comparing the visual narratives that circulate across platforms and topics. It can be applied to CCPP research areas such as misinformation and disinformation, propaganda, and imagery associated with political violence, helping scholars systematically analyze how these issues are visually represented and framed within media and online ecosystems.