Roth,S., Mattis,J.

SAGE (a system for automatic and graphical explanation)

Roth and Mattis (1990), in their SAGE system, extend Mackinlay's work (APT) in another way, adding more semantic description to the data and their relationships. SAGE is also based on the generate-and-test paradigm. It uses numerous variations and syntheses of 2-D static displays found in business and statistical graphics packages (e.g. bar and and plot charts, node-link graphs, gauges, techniques using shape, color or size, tables). The types of information with which this system is concerned are the relations among data values contained in relational database or frame.based representations.

The system uses several criteria by which it judges the relevance of different data characteristics. To choose an appropriate graphical technique SAGE uses several characteristics of a set which are Set-Ordering, Coordinates Vs. Amounts, Domain of Membership.

Set-Ordering. The nature of ordering relationship among a data set's elements is the predominant criteria used in APT and one criteria in SAGE for choosing graphical techniques. An ordering technique can be characterized as either quantitative, ordinal or nominal.
Coordinates Vs. Amounts. SAGE can recognize that elements of ordered sets are coordinates if each element specifies a point or location temporally, spatially, or otherwise (e.g. calender-date, latitude). In contrast, amounts are not embedded in particular frames of reference (e.g.number of days, weight).
Domain of Membership. SAGE's characterization goes beyond APT's by recognizing that sets can belong to the different domains of time, space temperature, or mass. This information helps to preserve subtle stylistic conventions, such as using a horizontal axis for time coordinates and a vertical axis to temperature.

For SAGE three properties are defined which describe the way relations made from elements of one set to another:
Relational Coverage. This characteristic describes whether every element of a set can be mapped to at least one element of another.
Cardinality. This characteristic expresses the number of elements of a set to which a relation can map from an element of another set.
Uniqueness. This characteristic refers to whether a relation maps to a unique value(s) for each element of a set.
Algebraic Dependencies. Algebraic dependencies among database elements suggest another dimension which can affect presentation design. Dependencies can occur among attributes (relations) or among values within data sets. For example, an organizational database may contain three relations mapping departments to dollar-amounts: materials-costs, labor-costs, and total-costs, where total-costs= materials-costs + labor-costs.

One of the most important issues for graphical design not addressed in APT is the role of an application or user's goals in viewing data. Differences in goals can greatly alter the effectiveness of graphical techniques or their combinations. A user's immediate goal may determine the connection of different relations and thereby affect how they should be integrated in the presentation. SAGE uses a technique of segmenting the total presentation request into sub-requests to coordinate text and graphics displays by converting a topic outline prepared by a discourse processor into a serial list of sub-requests. This indicates to the graphics system that information expressed in contiguous portions of the text should be considered more related, and therefore displayed together.
More generally, this characteristic provides a vehicle for expressing two, often competing, informational goals: the need to express as much information as possible and the need to selected partitions of sets or relations to be easily and cohesively viewed.


Visualization Concepts

Last modified on March 29, 1999, G. Scott Owen, owen@siggraph.org