Aggregating community maps.

This paper is motivated by a practical problem: many U.S. states have public hearings on "communities of interest" as part of their redistricting process, but no state has as yet adopted a concrete method of spatializing and aggregating community maps in order to take them into account in the drawing of new boundaries for electoral districts. Below, we describe a year-long project that collected and synthesized thousands of community maps through partnerships with grassroots organizations and/or government offices. The submissions were then aggregated by geographical clustering with a modified Hausdorff distance; then, the text from the narrative submissions was classified with semantic labels so that short runs of a Markov chain could be used to form semantic sub-clusters. The resulting dataset is publicly available, including the raw data of submitted community maps as well as post-processed community clusters and a scoring system for measuring how well districting plans respect the clusters. We provide a discussion of the strengths and weaknesses of this methodology and conclude with proposed directions for future work.