An Interactive Machine Learning System for Image Advertisements

Abstract

Advertising is omnipresent in all countries around the world and has a strong influence on consumer behavior. Given that advertisements aim to be memorable, attract attention and convey the intended information in a limited space, it seems striking that previous research in economics and management has mostly neglected the content and style of actual advertisements and their evolution over time. With this in mind, we collected more than one million print advertisements from the English-language weekly news magazine “The Economist” from 1843 to 2014. However, there is a lack of interactive intelligent systems capable of processing such a vast amount of image data and allowing users to automatically and manually add metadata, explore images, find and test assertions, and use machine learning techniques they did not have access to before. Inspired by the research field of interactive machine learning, we propose such a system that enables domain experts like marketing scholars to process and analyze this huge collection of image advertisements.