Essays on Visual Marketing

Abstract

As firms are embracing visual platforms in their marketing and branding efforts, little research exists on the relative effectiveness of visual versus text-based marketing efforts. In essays 1 and 2, we develop a quantitative framework to study how text and visual components of firm communications affect consumer engagement with firm-generated social media content. First, we quantify the emotional loading of text and imagery on sentiment and arousal/motivation-to-act dimensions. We use existing NLP tools for text and use machine learning and computer vision to develop and train sentiment and arousal classification models for imagery. We use four emotion modalities as predictors of visual emotion: (1) elements of design (low-level visual features) such as color, texture, shape, lines, curves, corners, edges, and orientation; (2) high-level visual objects/concepts (e.g., adventure, action, leisure, danger, etc.); (3) human facial expressions; and (4) text embedded in the image. We find that elements of design and high-level visual objects are the most important predictors of visual emotion. Our model achieves accuracy over 80%. Next, we apply the procedure to an empirical analysis of engagement (retweeting) with firm-generated content based on 1.3M tweets of 600+ brands from 11 categories, posted since 2008. Our findings suggest that over the years, consumers have developed resistance to persuasion messaging using high motivation-to-act text. We do not find a similar decline in effectiveness for high motivation-to-act imagery. We find significant heterogeneity of image effects by industry, with positive and high motivation-to-act imagery being the most engaging for quick-service restaurants, and negative imagery being the most engaging for charities/non-profits. In essay 3, we study the effects of the face and gaze of models in the product images on outcomes such as product clicks, orders, and returns. We use deep-learning algorithms for face detection and gaze-following in the context of 57,088 apparel products from 22 categories on a large Chinese e-commerce website. We find that product images that include the model’s face receive more clicks. However, higher prominence of the face leads to fewer clicks, but more ordering of the product. We also observe that the “direct” gaze and “downwards” gaze of the models in the images lead to fewer clicks and orders than “sideways” gazes. We offer potential explanations based on gaze psychology.