Picture perfect: An image mining of advertising content and its effects on social targeting

Abstract

Using a Google's image analysis framework that is based on machine learning algorithms, this study investigated how diverse image components of advertisements designed for fashion products affect consumers' search behaviors and purchase propensities in mobile environments. We focused particularly on the impact of various image attributes on the effectiveness of two popular forms of social targeting, namely the dynamic product advertisement (DPA) and customer audience (CA) strategies. We obtained tailored advertising image contents for 13,040 products across 84 different retailers through random sampling and extracted five image content features: (1) facial expressions, (2) degrees of stimulus content, (3) image classification, (4) color properties, and (5) texts contained within the images. We then examined how such ad image attributes affect consumers' click-through rates and conversions and determine the effectiveness of social targeting on the basis of simultaneous estimation frameworks. Our results revealed that emotional appeal (i.e., sexual images) positively influences consumers who opt for peripheral routes (e.g., click-through situations), whereas informational appeal (i.e., text-embedded images) is effective for people who select central routes to process a high cognitive load (e.g., purchases). We found that semantic and color complexities exert varying effects on consumers' search and purchase behaviors. We validated that changes in attributes toward marketing stimuli vary in congruence with a user's prior knowledge. Overall our study lay a foundation on the effects of image-based social targeting on consumer behaviors that are complex and dynamically evolving in digital, mobile environments.