dc.description.abstract | The visualization field has been part of the advancement of science throughout
history. In the 1970s, a movement emerged that advocated for graphical
representations as one of the necessary pillars of data analysis. Since then,
numerous scientific fields and businesses have adopted visualization as a
pillar to facilitate understanding information and disseminating results. Currently,
part of the research in this field focuses on new approaches aimed at
increasing the accessibility of visualization in order to promote its adoption.
These works maintain a balance between visualization expressiveness and
ease of use in order to address the needs of the broadest audience possible.
However, they remain primarily limited to developing single views, a highly
limiting factor to their adoption. Complete data analysis requires a mix of
different visualizations and data processing algorithms. Furthermore, to
reduce the complexity problem inherent to visualizations and to maximize
the analysis capabilities, it is necessary to use highly interactive visualizations
and multiple views.
Some state-of-the-art works define visualization dashboards that implement
this multi-view paradigm, allowing their coordination through graphical
user interfaces. To further facilitate this perspective, other methods
use data flow diagrams tailored for visualization to assemble the functionality
that drives these visualization dashboards. These diagrams have the
advantage of being a visual programming approach, which makes them
very accessible to users without programming skills, further increasing the
accessibility of these tools. However, the field of visualization, and science in
general, is a rapidly changing field where researchers are constantly proposing
new advances. Therefore, the previous tools must be able to adapt to new developments in order to maintain pace with this constant evolution.
Despite this, state-of-the-art visualization solutions are usually highly limited
to a fixed functionality.
This thesis addresses the previous problems by defining a meta-framework
for building data analysis applications based on coordinated and multiple
views through data flow diagrams, VMetaFlow. Unlike other works that
follow this approach, we use diagram connections to model coordination.
Therefore, our approach allows us to extend the range of supported interactions
between views, effectively supporting a more comprehensive range
of multiple views techniques. Furthermore, we define our meta-framework
as an abstraction layer over concrete visualizations and data processing
algorithms implementations. This way, our work can adapt and benefit from
new advances in these fields by integrating the latest developments.
Finally, to evaluate our work, we define three use cases and a critical
inspection with expert users. The first use case allows us to compare
VMetaFlow with the most recent work based on data flow diagrams for
building visualization applications. In the second use case, we address an exploratory
analysis where combining and coordinating information visualization,
scientific visualization, and data processing modules is necessary. In the
third use case, we address the study of the behavior of large-scale optimization
algorithms using visualization and dimensionality reduction, research
that we will publish shortly. Finally, the visualization experts interviewed for
the critical inspection confirm the advantages of our meta-framework over
other state-of-the-art works. | es |