Burning the Haystack to Save the Needle: Dimension Reduction and Privacy in Text and Network Data

Abstract

This poster presents results from applying a new dimension reduction technique (UMAP) to a wide variety of data types, ranging from online text to social networks, for the purpose of creating useful, but anonymized, data. As the dimension reduction procedure produces meaningful distances and supports arbitrary distance measures, it can be applied to a variety of problems, and produces data that is useful for both visualization and predictive modeling. Included is a description of the dimension reduction procedure, the results of its application, and a discussion of planned future use.