| We define Table Visualizations (TVs) as visualizations on tables of data, typically m x n, where n is the number of dimensions or columns and m is the number of records or rows. The dimensions or features can be of various data types: ordinal, categorical or nominal. This data model is typical of many databases and these visualizations can also be called multidimensional multivariate information visualizations, since inherent spatial dimensions may not be present. Many visualizations such as scatter plots, bar charts, histograms, and line plots are TVs, and usually represent only two or three dimensions. Scatter plot grids, parallel coordinates, survey plots, iconographics, n-dimensional grand tour projections, circle segments, dimensional stacking, radviz and other advanced visualizations are also TVs designed to handle a larger number of dimensions. In this research we analyzed and evaluated many Table Visualizations. We created objective and subjective methods measuring specific desired data mining features, including a display utilization grid, a mechanism for producing statistical values for a given visualization. We analyzed and parameterized a set of specific TVs and defined basic TV operations. We introduce dimensional anchors, a graphic primitive that encodes these basic visualization operations. Dimensional anchors can describe a large subset of TVs and thus produce a taxonomy of visualizations. We defined a space of TVs and used this space to create new visualizations, to interpolate between visualizations and to take Grand Tours of visualization spaces. |
