Ramana Rao and Tichomir Tenev
InXight Siftware and Xerox PARC
3400 Hillview Avenue; Palo Alto VA 94304; USA
{rao, tenev}@inxight.com
Many datasets common and business (and elsewhere) are very large multidimensional tables with some dimensions having hierarchical structure (e.g. Years & Quarters, Regions & Sales people, Channels, Products, Line Items). Often the business user would like to analyze and browse they got interactive lead by Selecting and arranging dimensional categories and not in terms of relational storage models and report generation. To address this need, many multidimensional data bases and OLAP (On-Line Analystical Processing) products have been introduced over the last five or eight years. OLAP systems typically allow exploring multidimensional business datasets using a series of intuitive user interface actions involving dragging around tiles that represent rows, columns, or dimensions to perform "slicing and dicing" and "drilling down" and "rolling up" operations.
The Table Lens [1] is a focus plus context technique that allows visualizing and manipulating large 2-D cases by variables tables. Because it display is much more of the table at once, a user can examine patterns in the whole table as well as zoom in on specific contents without losing global context. We have extended the Table Lens visualization techniques to deal effectively with large multidimensional datasets and to support typical OLAP operations [2].
Typical OLAP systems are based on textual tables. This limits the number of actual values that can be displayed at once as well as makes it hard for the user to perceive patterns or outliers in the data. while Table Lens uses graphical representations to both compress values and map the attentive acts of interpreting textual symbols into equivalent perceptual acts of comparing graphical properties. A related problem is that graphical views of data are obtained through a process of selecting data and generating graphical views using dialogs to control choices. The Table Lens in beds in the corresponding graphical representation of values directly into the table simplifying the process of generating the most common kind of graphics and providing a useful point of departure for most complex presentation of graphics.
Our recent work extends the Table Lens framework to support OLAP operations. The traditional Table Lens view is equivalent to relational encoding of the multidimensional table. Each of the dimensions is modeled as an additional column with each of its items being one possible value of the column domain. Because of the regular structure of the multidimensional table, the values in the columns will seem to repeat with a periodicity depending on the ordering of dimensions as columns and row sorts. this layout of data tends to produce a very large tables as well as tables with when populated section, which leads to various refinements (including multi-level focus) on Table Lens design. The new operations enable starting with the traditional view and proceeding to a multidimensional view of using a tile dragging paradigm.
The integration of OLAP operations into the graphical and focus+context framework allows the user to navigate around multidimensional datasets and focus in on subsections opportunistically. in addition, the table can be rearranged and manipulated and summary views of selected dimensions and aggregates groups can be produced quickly as an integral part of the exploration.
References
[1] Ramana Rao and Stuart K Card, The Table Lens: Merging Graphical and Symbolic Representations in an Interactive Focus+Context Visualization for Tabular Information, Proceedings of the ACM SIGCHI Conference on Human Factors in Computing Systems, Boston, MA, April 1994, ACM.
[2] Tichomir Tenev, SpreadCube: A Visualization Tool for Exploratory Data Analysis, Masters Thesis, Massachusetts Institute of Technology, January 1997.