Advertising over Time: Interactive Machine Learning for Image Data Analysis

广告在世界各地的各个社会无处不在，对人类的行为有很大的影响。如今，L、辉瑞、宝洁和卡夫亨氏等公司每年在杂志的平面广告上总共花费20多亿美元。过去的研究表明，广告对销售、品牌忠诚度、市场份额稳定性等有不同的影响。然而，除了广告支出的水平，广告的实际内容和风格也是至关重要的。现有的经济学和管理学文献普遍忽视了这一点。随着大数据和当今日益增长的计算能力的到来，我们有了一个随时可以填补这一研究空白的推动者。更准确地说，从1843年到现在，我们拥有一套超过100万份的平面广告。我们建议应用机器学习和人工智能技术来提取广告中的相关特征，并使用它们来识别广告内容和风格中的模式。我们的研究项目包括两个主要支柱：一个是通过开发一个“交互式机器学习系统”来获取设计知识。该系统由两个部分组成：“交互式标注”和“交互式分析”。互动的目的是创建一个直接让用户参与学习和分析过程的系统，并旨在满足用户的需求。我们建议的研究项目的另一个支柱是使用开发的系统在广告研究中产生新的见解。我们将回答的问题是：平面广告随着时间的推移发生了怎样的变化，特别是在广告文本、广告图像特征和广告复杂性方面？这些变化与提高消费者生活速度的外部事件有多大关联？

Advertising is omnipresent in all societies around the world and has a strong influence on human behavior. Today, companies such as L'Oréal, Pfizer, Procter & Gamble and Kraft Heinz together spend more than $2 billion a year on print ads in magazines. Past research has shown various effects of advertising (e.g. on sales, brand loyalty, market-share stability etc.). However, in addition to the level of advertising spending, the actual content and style of advertising is also crucial. This is something that the existing economics and management literature has widely neglected.With big data and todays increasing computing power at hand we have an enabler ready to fill this research gap. To be more precise, we possess a set of more than 1 Mio print ads from 1843 up to today. We propose to apply Machine Learning and Artificial Intelligence techniques to extract relevant features from the ads and use them to identify patterns in the content and style of the advertisements.Our research project consists of two main pillars: One of them is to derive design knowledge from the development of an "Interactive Machine Learning System". This system consists of two components: "Interactive Labeling" and "Interactive Analytics". The interaction is intended to create a system that directly involves the user in the learning and analysis process and is designed to meet the user's needs. The other pillar of our proposed research project consists of using the developed system to generate new insights in advertising research. Questions that we will answer are: How have print advertisements changed over time, especially with regard to ad text, ad image characteristics and ad complexity? How far are these changes related to external events that increase consumers' speed of life?