Abstract

Photos play a critical role in online shopping. To examine their impact on consumers, most previous studies rely on human assessments to develop measures for photos. Such an approach limits the number of dimensions and samples that can be investigated in one study. This study exploits image-processing techniques to tackle this challenge. We develop a framework and differentiate two types of computer-generated measures, aggregative and decompositive measures, which may be used in different ways in empirical research. We review the major image-processing technologies that have potential to be used in consumer behavior research. To showcase the feasibility of the framework, we conduct an example study on product photos’ impact on consumer click-through. Moreover, we conduct a simulation to investigate the robustness of the framework under the attack of image-processing algorithm errors. We find that image-processing techniques with 90~95% accuracy will be sufficient for empirical research.

中文翻译：

---

利用图像处理技术进行实证研究：在线购物环境中的可行性和可靠性

摘要

照片在网上购物中起着至关重要的作用。为了检查它们对消费者的影响，以前的大多数研究都依靠人工评估来制定照片测量标准。这种方法限制了一项研究中可以研究的尺寸和样本数量。这项研究利用图像处理技术来应对这一挑战。我们开发了一个框架，并区分了两种计算机生成的量度，即汇总和分解量度，它们可以在实证研究中以不同的方式使用。我们回顾了可能在消费者行为研究中使用的主要图像处理技术。为了展示该框架的可行性，我们对产品照片对消费者点击率的影响进行了示例研究。此外，我们进行了仿真，以研究在图像处理算法错误的攻击下框架的鲁棒性。我们发现，具有90〜95％精度的图像处理技术将足以进行经验研究。