Ian Taylor and Steve Benford
Department of Computer Science
The University of Nottingham
Nottingham. NG7 2RD, UK
{i.taylor, s.benford}@cs.nott.ac.uk
http://www.crg.cs.nott.ac.uk
Introduction
Various systems have demonstrated alternative approaches to abstract three-dimensional representations of multi-dimensional datasets. Examples of these include VR-VIBE[1], QPIT[2,3], SDM[4] and the Open Text Web Index. This second of these approaches uses of this. Our plot technique described by Michael Benedikt[6] in which graphical attributes are directly mapped to data attributes to form the image. This has shown to be particularly effective in the observation of trends and anomalies in large datasets as it exploits the human visual system's ability to isolate groups of points based on a position, color size and shape [7]. This poster describes a three-dimensional scatter plot system in which graphical attributes may be dynamically mapped to data attributes and image re-generation performed instantly. Seven graphical dimensions are available which may be assigned to numeric, alphanumeric or enumerated field types. Geographic placement is also offered and the scatter geographic plots may be switched between instantly in the same dataspace.
The system goes further to allow representation of the data hierarchy. Graphical objects can themselves represent tuples which point to any number of tables. These can be selected to instantly spawn new visualizations of the sub-structure to which they refer. The definition of tables and references is the same at any level and so no constraints are placed on the number of levels in the hierarchy. Key references between relation tables may be displayed by graphical links, as can any network structure.
Figure 1 shows an example single table Benediktine plots and Figure 2 demonstrates the geographic display. Once drawn the objects may be subject to certain manipulative operations including filtering and re-sizing. Figure 3 shows a single table scatter plot subject to a combination of filtering options.
Database Application
An important part of this work was in the application of the system to a commercial datasets provided by industrial sponsors. The ease of use and versatility of the visualization enabled hitherto un-noticed trends to be revealed in the behavior of certain variables. This was greatly facilitated by the ability to construct the image intuitively by exchange and addition of graphical dimensions. Images produced as a result of this exercise can be seen in figures 4, 5 and 6. The process is explained and resulting criticisms of the system are included.
The system has also been applied to a small relation all system and an example plot can be seen in Figure 7. In this graph each object represents a table and key references are shown by the directional links. The plot is subject to the same dynamic placement and manipulation techniques as a single table. Any object in the graph may be expanded to spawn a scatter-graph of the single relation. Figure 8 shows two visualizations which were spawned from the database overview image (also visible).
This work is continuing to examine the suitability of this approach to a more complete framework for the visualization of the relational model. Part of this is to construct methods by which the operations provided by relational algebra may be incorporated into the visual environment. Topics such as multi-user interaction, user tracking, preservation of semantics and knowledge sharing are under construction in the context of these techniques. Additional tools for visual interface (e.g. [8][9]) are under evaluation as potential enhancements.
References
[1] Benford, S., Snowdon, D., Greenalgh, C., Ingram, R., and Knox, VR-VIBE: A Virtual Environment for Co-operative Information Retrieval, Proc. Eutographics '95, Maastrict, The Netherlands, August 1995.
[2] Steve Benford, David Snowdon and John Mariana, Populated Information Terrains: first steps. In R. A. Earnshaw, J. A. Vince, H. Jones (ed), Virtual Reality Applications, pp 27-39. Academic Press, 1995.
[3] http://www.comp.lancs.ac.uk/computing/research/cseg/projects/qpit/
[4] Mei C. Chuah, Steven F. Roth, Jow Mattis, John Kolojejchick. SDM: Malleable Information Graphics. Proceedings of Information Visualization '95, Atlanta, USA, Oct 30-31, 1995, pp. 36-42.
[5] Tim Bray. Measuring the Web. Proceedings of Fifth International World Wide Web Conference, Paris, France, May 6-10, 1996, pp. 993-1005.
[6] Benedikt, M. (1991). Cyberspace: Some proposals. In M. Benedikt (ed.) Cyberspace: The first steps. Cambridge, MA: MIT Press.
[7] Csinger, Andrew, The Psychology of Visualization (Technical Report), no. TR-92-28, University of British Columbia, Nov 1992.
[8] Maureen C. Stone, Ken Fishkin, Eric A. Bier. The Movable Filter as a User Interface Tool. Proceedings of CHI '94, Boston, MA, USA, April 24-28, 1994, pp. 306-312.
[9] Rob Ingram and Steve Benford, Legibility Enhancement for Information Visualization, Proceedings of Visualization '95, Atlanta, USA, Oct 29-Nov 3, pp. 209-216.