Database Reference
In-Depth Information
readability of a visualization, but introduce errors in the data themselves. Data may
have uncertainty attributes associated with them, which can interfere with the meas-
urement. On the other hand, there are numerous examples of improving information
content by using novel layout, shape, and color strategies or augmenting the visualiza-
tion with links, boundaries, and even white space. The amount of information con-
tained may also be enhanced using redundant mappings, which improves the chances
of successful reception by the viewer. Finally, the use of animation to communicate
information in an incremental fashion can be quite effective; it is lossy communica-
tion, as viewers quickly forget some of what they have seen, but the ability to replay
the animation can replace some of this lost information.
3.6
Case Study: Parallel Coordinates
As an example of this information measurement activity, let us consider parallel coor-
dinates, a popular multivariate visualization technique. The first question is how well
does this technique present the values of a dataset? For individual values, this method
has very high resolution, given most of the screen height can be used to convey the
value. This implies the technique possesses high selective information content, at least
on individual dimensions, as separation into sub-ranges is facilitated by the amount of
screen space allocated to convey values. However, the loss due to occlusion can be
high, especially for nominal variables. This loss is somewhat mitigated by the varying
slopes of lines ending/starting at axes, which allows some degree of differentiation.
Relationships among data along an axis can be emphasized via embedded histograms,
as found in some implementations of parallel coordinates. In terms of relationships
between dimensions, this method is limited to showing pairwise relations, with N-1
out of the N*(N-1)/2 possible relations shown. Automated dimension ordering can
help reveal interesting relationships, as seen in [14]. Relationships between records
are problematic due to the ambiguous continuity of records that intersect on one or
more axes. Coloring the lines based on a record ID can help with modest sized data-
sets. We can also use animation along an axis or based on an order to expose some
inter-record relations. Intersections and near-parallel edges can reveal partial struc-
tures (between dimension pairs), and techniques such as Hierarchical Parallel Coordi-
nates [17] can show grouping relations, though there is loss of individual data values.
Each of these methods enhances the descriptive information content of the visualiza-
tion, thus helping analysts form mental models of the data.
Through analyzing these augmentations of parallel coordinates we see that many
recent innovations to parallel coordinates target different types of information loss
resulting from this means of mapping data. For example, we see efforts to preserve
and emphasize outliers in a paper by Novotny and Hauser [18]. However, many other
issues still exist; there are still many data features that cannot be readily perceived and
tasks that are difficult to perform using parallel coordinates.
3.7
Conclusions
To summarize, we feel there are many aspects of information visualization research
that can find analogies in the concepts of information theory - it is all about commu-
nication. Perhaps finding such a formal structure on which to ground our efforts can
Search WWH ::
Custom Search