Theoretical Foundations of Information Visualization - Information Visualization: Human-Centered Issues and Perspectives - page 49

crease and decrease are examples of patterns. A model may be a synthesis of several
patterns each representing some part or aspect of the data. Thus, when the observation
of the morning temperatures is performed over a sufficiently long period, the model
will probably incorporate the patterns of both increase and decrease of the tempera-
ture. Furthermore, patterns may also be composed of sub-patterns. For instance, the
behavior of the temperature may be conceptualized as a repeated “wave” where in-
crease is followed by decrease. Here, increase and decrease are basic, or atomic, pat-
terns, the “wave” is a composite pattern including the increase and decrease patterns,
and the repetition of the “wave” is a pattern of a yet higher level, which incorporates
the “wave” pattern.
The main role of Information Visualization tools can be understood as helping the
user to perceive patterns that could be used for building an appropriate model. This
means, in particular, that a tool should facilitate the perception of (sub)sets of data
items as units. For an appropriate support of the detection of patterns, a tool designer
should know in advance what types of patterns need to be perceived (or otherwise
detected) with the use of the tool. Then, after the tool is ready, it will be easy to ex-
plain to the users the purpose of the tool and instruct them how to detect the types of
patterns the tool is oriented to.
The types of patterns that may be meaningful for the user depend on the structure
and properties of the data under analysis. Thus, in the analysis of a temporal series of
numeric measurements (such as temperatures) it makes sense to look for such basic
patterns as increase, decrease, stability, fluctuation, peak, and low point. However,
when numeric measurements refer to a discrete unordered set as, for example, melting
temperatures of various substances, the possible types of patterns may be groupings of
elements with close values of the measurements and frequency-related patterns:
prevalence of certain values or value intervals, frequent values or exceptional values
(outliers).
To support the designers and users of Information Visualization tools in the way
described above, there is a need for a theory that could enable the possibility to pre-
dict, for a given dataset or a given class of datasets, what types of patterns may be
found there. We specially emphasize the term types to exclude the possible impres-
sion of attempting to predict (and on this basis automatically detect) all specific pat-
terns hidden in specific data. Thus, a prediction that a dataset may contain groups
(clusters) of objects with similar characteristics does not define what specific clusters
are there. However, it orients tool designers, who will know that the tool must help
the users to detect clusters, and users, who will know that they need a tool facilitating
the detection of clusters. Then, if each Information Visualization tool and technique is
supplied with an appropriate signature (i.e. what kind of data it is suitable for and
what types of patterns it is oriented to), the user will be able to choose the right tool.
The theory we are advocating in this section can be called data-centered predictive
theory. The theory needs to include

1. an appropriate generic framework for the characterization of various data

types and structures;

2. a general typology of patterns;

3. a mechanism for deriving possible pattern types from data characterizations.

Here, we present some preliminary ideas concerning these components of the theory.