More than Distant Viewing: Qualitative Views on Machine Learning as an Automated Analysis Method in Networked Climate Image Communication

Machine learning algorithms are increasingly used within the digital humanities as heuristic methods in image analysis. Automated image recognition methods and the sorting of image similarities provide access to large image inventories, which are made accessible to a human eye primarily via visual distance methods. In this context, the visualisations as an interactive interface represent only one 'end product' of a dense series of both qualitative and quantitative methodological decisions within a research design. Especially these decision-making processes are the crucial points within DH research, since with them the boundaries of qualitative-image-scientific and quantitative-algorithmic approaches become perceptible. This article addresses the process of a semi-automated analysis procedure of large image data sets using Google Images on climate change as an example. It results from a case study on cross-cultural climate image comparison by the mixed-methods research team anci. https://anci.fh-potsdam.de/ With the help of t-SNE as a machine learning method for dimensionality reduction and the k-means clustering method, the methodological process from data visualisation to visualization is made critically and reflectively transparent. The term pipeline serves as a metaphor for the methodological process.