Shape Search in Temporal Data to Facilitate Knowledge Discovery: A User Interface to Find Spikes, Sinks and Slopes

The ability to find interesting patterns in sequential datasets is important in both data analysis and knowledge discovery. Shapes, such as spikes, valleys, and increasing lines, created when graphing sequential data points are familiar to analysts as a way of identifying trends and anomalous behaviors. This work presents a set of common shapes that can be used by visualization designers to assist users in discovering patterns in data that may be interesting. Each shape has a set of characteristics that can be used to rank their “interestingness.” A way of identifying each shape and the characteristics used to rank them are presented in this paper. Our method of shape identification and ranking was integrated into FeatureLens, a user interface to explore and visualize features in collections of text documents. A case study was conducted with a literary scholar to analyze patterns of repetition in a modernist novel.