Dados Bibliográficos

Resumo

This paper presents a practical guide to machine learning–assisted visual content analysis for social scientists. Combining machine automation with human expertise and reflexivity, the proposed methodological framework bridges the gap between computer vision and social research. Our custom approach combines inductive, deductive, and abductive logics of scientific inquiry and consists of three complementary steps: (a) Pattern exploration—employing unsupervised learning to explore visual patterns within image datasets; (b) Theory-driven image classification—utilizing supervised learning with convolutional neural networks to systematically label visual content; and (c) Context-sensitive interpretation—to provide critical and creative engagement with the patterns identified in the previous steps. We illustrate these three steps, and their various combinations, through empirical examples from a study of visuality in digital diplomacy, and critically discuss the epistemological implications of using machine learning as a method in visual social research.

Referências Citadas