Nonlinear dimensionality reduction as sematic basis of data visualizations
Similarity visualizations are an increasingly central element of data exploration in digital humanities and cultural heritage research. This makes it important to understand what such meta images of big image data show, what they hide, and what they make believe. In order to investigate those rather new visualization techniques diagrammatically, a mathematical understanding, a technical experimentation, a theoretical analysis as well as a historical contextualization is needed.
This R-Code uses open data from the Museum of Modern Art, takes metadata as features, reduces the dimensions with the t-SNE algorithm and displays the images as an image plot. The goal is to create an environment in which one can experimentally examine the semiotic structure of t-SNE-based meta-images. By varying individual factors, their semantic value can be explored in their application on large image databases.
