Analyzing Image Data with Machine Learning

Visual information such as images and videos is abundant in the digital age. However, visual data are currently underutilized in social science research. Previous sociological studies that rely on visual information as data either use qualitative interpretation or small-scale quantitative analysis, which is not suitable for large-scale datasets. This chapter first provides a historical overview of automated image analysis using computer vision. It then discusses the state-of-art deep learning algorithms and summarizes the two steps of using deep learning algorithms to automatically analyze large image datasets. It uses various publications using large-scale image datasets ranging from social media images, satellite and street view images, and historical archives to illustrate how automatic image analysis was carried out in recent research. Finally, the chapter discusses ways in which social scientists can better utilize visual information in their future research.

2