Visualizing the quality of dimensionality reduction

The growing number of dimensionality reduction methods available for data visualization has recently inspired the development of formal measures to evaluate the resulting low-dimensional representation independently from the methods' inherent criteria. Many evaluation measures can be summarized based on the co-ranking matrix. In this work, we analyze the characteristics of the co-ranking framework, focusing on interpretability and controllability in evaluation scenarios where a fine-grained assessment of a given visualization is desired. We extend the framework in two ways: (i) we propose how to link the evaluation to point-wise quality measures which can be used directly to augment the evaluated visualization and highlight erroneous regions; (ii) we improve the parameterization of the quality measure to offer more direct control over the evaluation's focus, and thus help the user to investigate more specific characteristics of the visualization.