Network-based analysis for identifying visual patterns in advertising

Abstract

Digital advertising played a pivotal role in shaping public perceptions of technology, with Instagram emerging as a predominant platform for visual brand communication. In this context, the analysis of visual patterns emerged as a strategic approach to comprehending the narratives that underpinned the positioning of technology companies. This article investigated visual patterns in the digital advertising of technology brands by constructing and analyzing similarity networks among images published on Instagram. A total of 4,580 images from four brands (three Brazilian and one multinational) were collected, and these were selected based on objective market presence and economic sector criteria. The images were transformed into embeddings using convolutional neural networks. A graph was then constructed and analyzed using the Gephi software, with the analysis based on vector similarity. Modularity and centrality metrics were applied to identify visual structures. The results of the analysis revealed the presence of cohesive clusters, each exhibiting a distinct graphic style. Among the brands analyzed, Positivo Tecnologia demonstrated a high degree of visual cohesion, characterized by the clear delineation of thematic groupings and a discernible strategic organization. Centrality metrics identified influential images within clusters, while modularity scores highlighted the fragmentation and centrality of visual concepts. The efficacy of the embedding-based and network analysis approach in mapping visual patterns in digital advertising was demonstrated, thereby revealing the aesthetic coherence and visual identity of technology brands. The adopted methodology delineated a replicable paradigm for prospective investigations into digital communication strategies, thereby contributing to advancements in the domains of communication design and computational image analysis.