Image content effectiveness analysis of social media posts using machine learning methods
Date
2021
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
With the evolution of the Internet, social media sites nowadays have become increasingly important in marketing campaigns for brands and companies. And thanks to faster Wi-Fi and cellular speed and better phone cameras, more and more images are being created and shared on social media every day. As more visual contents appear, understanding how consumers respond to visual content is essential to the brand’s social media marketing strategy. ☐ Rather than merely checking the effect of the presence of the images in the post, our research goes more in-depth and aims to shed light on the role of different image characteristics. In this research, we evaluate images along two dimensions – informational value and entertaining value. We use whether the image features a product to measure the informational value and use aesthetic appraisal and perceived emotional arousal to measure the entertaining value, where the aesthetic appraisal indicates how beautiful the viewers feel about the image and the emotional arousal indicates the level of evoked emotionality in the viewers. To carry out the empirical analysis, we collect 3,803 brand posts containing 10,259 images from 8 international brands. We bring up a framework for image content analysis, in which computer vision techniques and image classification algorithms are utilized to investigate the image content features. Then we build a regression model to study how image’s informational value and entertaining value affect customer engagement with the brand post, including the moderating effects of different themes, along with other image and post features. The results highlight the importance of finding the right pictures with proper features for posts with different themes. We also identify 17 interpretable image attributes across five dimensions relevant to the image’s aesthetics and arousal to give direct and applicable implications for optimizing images on social media platforms. ☐ Our key findings are as follows. (1) An image’s entertaining value is more important than its informational value in driving customer engagement; (2) It is important to consider the compatibility between image content features and the theme of the post; and (3) Aesthetic appraisal and emotional arousal, the two entertaining value measures, are related to different objective image attributes. ☐ By analyzing how those image contents affect social media audiences, our results help marketing practitioners better design the appropriate images and capitalize on social media engagement.
Description
Keywords
Computer vision, Deep learning, Image content mining, Machine learning, Social media, Visual marketing