LEARNING VISUAL ATTRIBUTES FOR DISCOVERY OF ACTIONABLE MEDIA

Abstract

Since the advent of social media, it became common practice for marketers to browse user generated content in social networks sites to discover actionable media that resonate with the brand identity and have the potential of engaging a target audience. However, there is still no concrete understanding of what are the discriminative visual attributes of actionable media. We formalize the task of discovery of actionable brand media as a learning-to-rank framework and design a discovery framework that integrates three different classes of visual attributes. The method addresses the numerous challenges derived from the subtle differences between competitor brands, the sparsity of interactions between posts and brands, the diversity of each brand timeline and the subjective nature of visual actionability. Our comprehensive experiments and visualizations confirm that this work is a valuable concrete step toward using data-driven methods to discover actionable media for brands.