标题: VisBubbles: A Workflow-Driven Framework for Scientific Data Analysis [打印本页] 作者: 彬彬 时间: 2011-12-30 18:53 标题: VisBubbles: A Workflow-Driven Framework for Scientific Data Analysis 1 Introduction
Our long-term collaborations with bat flight biologists have revealed
that several workflow issues continue to be persistent barriers
in scientists’ data analysis tasks. Only limited attention has
been given to user interfaces that bridge programming and visualization
in environments involving multiple information resources,
feature extraction by programming, and analytic study. We present
VisBubbles (Fig. 1), a unified environment supporting programming,
visualization, and interaction concurrently for data analysis
workflow. The framework was developed in a participatory design
process and discussion with biologists continues to provide new insights
into general-purpose support for data-analysis workflow.
2 Design Analysis
Our long-term collaboration with bat biologists has revealed two
workflow design issues. The first is that visual interfaces have not
provided enough support in the dynamic data analysis process.
The biologists’ analytic process is dynamic, not static. That the path
of their exploration is often unpredictable imposes several design
requirements. They need easy access to a tremendous amount of
data to reason through their analysis, and they need an interface
that supports their multiple diverse and simultaneous tasks.
The second issue is that program implementation and data analysis
are conducted in multiple separate working environments.
Evolutionary biologists use Matlab to conduct analysis and then
program the results into visualizations for confirmation or generation
of new hypotheses. If more analysis is needed, they switch
back toMatlab to make changes. Switching back and forth between
Matlab and the visualization process introduces significant context
switching consts.
3 Our Methods and Preliminary Results
We addressed these issues in the design of VisBubbles, that has two
key design apsects. One key aspect of VisBubbles is its extension
of static multiple views to a metaphorical interface of bubbles that
becomes a flexible layout to support analysis, motivated by the recent
success of integrated development environment in Code Bubbles
[Bragdon et al. 2010]. Unlike windows, bubbles do not overlap
but instead push each other out of the way, thus supporting simultaneous
side-by-side comparison and freeing users from the windows
arrangement, so they can focus on their analytical tasks. Bubbles
can be moved freely in space to leverage the action sequence of
the data analysis process (considering humans’ limited mnemonic
abilities of 7 ± 2 items). Bubbles can also be grouped to form a
flexible multiple view environments. When grouped, data are automatically
linked to support interactive queries in such a way that
interacting in one view activates the same interactions in all other
grouped views.
e-mail: zgpan@cad.zju.edu.cn
†e-mail: jian.chen@usm.edu
1
2
3
5
6
4
8
7
9
11
12
10
Figure 1: VisBubbles interface. An example visual interface that a
biologist can build using VisBubbles, which contains (1) a virtual
work space and the following bubbles: (2) geometry, (3) compar-
ison plot, (4) parallel coordinates, (5) video, (6) note-taking, (7)
data source manager, (8) sketch bubbles for shape fitting, (9) web
bubbles, (10)Matlab programming bubbles, (11) 2D plots, and (12)
authorized data.
In a nutshell, biologists would look at all information to determine
which hypothesis is worth consideration. VisBubbles lets the users
load datasets of interest by composing the data (item 7 left) and
the good default visualizations (item 7 right) by direct dragging. A
new bubble pops up to show a visualization, e.g., as a mesh view
(Figure 1 (2)) or plots. By doing so, the interface supports presentation
of information of heterogeneous types. VisBubbles extends
the metaphor of bubbles [Bragdon et al. 2010] in that each bubble
represents a function unit that is valid for programming and can be
interpreted to create a visualization. Here each visualization is created
by a certain data type, e.g., spatial wing data are shown as a
three-dimensional (3D) mesh.
Another key aspect is the pipeline approach that combines programming
and interaction in such a way that newly authorized data can
be visualized without leaving the data analysis environment. Any
parameters derived from the Matlab can be placed in the authorized
data bubble (Figure 1(12)), which can also be composed with the
good default visualizations (Figure 1(7) right) to see newly authorized
data (Figure 1(11)). Biologists who used our interface reported
that our interface externalized the data analysis sequence by
presenting presentation, rather than hiding the query process, which
could aid more effective data analysis and student training.
References
BRAGDON, A., REISS, S., ZELEZNIK, R., KARUMURI, S., CHEUNG,
W., KAPLAN, J., COLEMAN, C., ADEPUTRA, F., AND
LAVIOLA JR, J. 2010. Code bubbles: rethinking the user
interface paradigm of integrated development environments.
ACM/IEEE International Conference on Software Engineering,