Query by Visual Words: Visual Search for Scatter Plot Visualizations

Finding interesting views in large collections of data visualizations, e.g., scatter plots, is challenging. Recently, ranking views based on heuristic quality measures has been proposed. However, quality measures may fail to reflect given user interest, since interestingness is strongly dependent on the application domain and user context. As an alternative, interactive exploration in combination with example based user queries can be used to find patterns of interest. We introduce a novel approach for searching in large sets of scatter plot views based on a dictionary of frequent local scatter plot patterns. The dictionary is used for interactive construction of scatter plot queries, taking into account similarity of local scatter plot patterns as well as their approximate location in the plot. We introduce the overall approach, present a glyph design for visualization of dictionary entries, and illustrate the applicability of our implementation.