Visualisation and Exploration of Scientific Data Using Graphs - Data Mining: Theory, Methodology, Techniques, and Applications - page 15

2. Able to access and integrate data from a number of sources. Data of interest
typically fall into one of three categories:
- databases within the AADC (e.g. biodiversity, automatic weather sta-
tions, and state of the environment reporting databases). These
databases are developed and maintained by the AADC, and so have
a consistent structure and are directly accessible.
- flat data files (including external remote sensed environmental data such
as sea ice concentration [8], data collected and held by individual scien-
tists, and data files held in the AADC that have not yet been migrated
into actively-maintained databases).
- web-accessible (external) databases. Several initiatives are under way
that will enable scientists to share data across the web (e.g. GBIF [9]).
3. Be web browser-based. A browser-based solution would allow the tool to be
integrated with the AADC's existing web pages, and thus allow clients to
explore the data sets before downloading. It would also allow any bandwidth-
intensive activities to be carried out at the server end, an important consid-
eration for scientists on Antarctic bases wishing to use the tool.
4. Have an intuitive graphical interface (suitable for a general audience) that
would also provide sucient flexibility for more advanced users (expected to
be mostly internal scientists).
5. Integrated with the existing AADC database structure. To allow the interface
to be as simple as possible, we needed to make use of the existing data
structures and environments in the AADC. For example, the AADC keeps a
data dictionary, which provides limited semantic information about AADC
data, including the measurement scale type (nominal, ordinal, interval, or
ratio) of a variable. This information would allow the application to make
informed processing decisions (such as which dissimilarity metric or measure
of central tendency to use for a particular variable) and thus minimise the
complexity of the interface.
A large number of software packages and algorithms for graph-based data
visualisation have been published, and a summary of a selection of graph software
is presented in Table 1 (an exhaustive review of all available graph software is
beyond the scope of this paper). Existing software that we were aware of met
some but not all of our requirements. The key feature that seemed to be missing
from available packages was the ability to construct a graph directly from a
data source (i.e. to create a graph that provides a graphical portrayal of the
information contained in a data source). Two notable exceptions are GGobi
[10] and Zoomgraph [11]. However, GGobi is intended as a general-purpose data
visualisation, and has relatively limited support for structured (nodes and edges)
graphs. Zoomgraph's graph construction is driven by scripting commands. For
our general audience, we desired that the graph construction be driven by a
graphical interface, and not require the user to have any knowledge of scripting
or database (e.g. SQL) commands.
This paper describes a prototype tool that implements the requirements listed
above. The key novelty of this tool is the ability to rapidly generate a graph