The emerging domain of data-enabled science necessitates development of algorithms and tools for knowledge discovery. Human interaction with data through well-constructed graphical representation can take special advantage of our visual ability to identify patterns. We develop a data visualization framework, called BiFold, for exploratory analysis of bipartite datasets that describe binary relationships between groups of objects. Typical data examples would include voting records, organizational memberships, and pairwise associations, or other binary datasets. BiFold provides a low dimensional embedding of data that represents similarity by visual nearness, analogous to Multidimensional Scaling (MDS). The unique and new feature of BiFold is its ability to simultaneously capture both within-group and between-group relationships among objects, enhancing knowledge discovery. We benchmark BiFold using the Southern Women Dataset, where social groups are now visually evident. We construct BiFold plots for two US voting datasets: For the presidential election outcomes since 1976, BiFold illustrates the evolving geopolitical structures that underlie these election results. For Senate congressional voting, BiFold identifies a partisan coordinate, separating senators into two parties while simultaneously visualizing a bipartisan-coalition coordinate which captures the ultimate fate of the bills (pass/fail). Finally, we consider a global cuisine dataset of the association between recipes and food ingredients. BiFold allows us to visually compare and contrast cuisines while also allowing identification of signature ingredients of individual cuisines.
展开▼
