How: Evaluation of visualization techniques for EDA: F C

1

• How Evaluation of visualization techniques for
  EDA FC
• Explain challenges of displaying large data set
  (how to see a million data points? ie. our age
  old screen space challenge)
• Explain the guideline of "overview, filter and
  detail-on-demand" in design. Relate this to EDA.
  For example, how to generate an overview if the
  data is not previously known to be hierarchical
  (ie. what to display, what to leave out if total
  number of data points exceeds the total available
  pixels). How does one provide detail-on-demand,
  PNZ, OD, FC? Is that task/data dependent?
• Explain how much do we know about the
  effectiveness and applicability of these
  visualization techniques, and how we contribute
  to existing knowledge by evaluating these
  techniques
• Theory
• General FC
• Evaluations lit review
• Text
• Graphical
• Evaluations real-life tasks
• Tree (Adam/Dmitry study CHI 2006 ??)
• Time-series data/xy data (LGE study)
• Evaluations human perceptual studies
• IT4
• IT5

• What Data
• Explain what large means using the Agilent high
  through-put instrument data, and Google keynote
  study clickstream data
• Explain that the data sets are not only large,
  but growing due to the relative ease in data
  collection
• Explain the data itself multivariate (?) with
  metadata (both continuous and categorical). Data
  may have hierarchical structure. In our
  examples, they are both time-series data. The
  instrument data is continuous (subject to
  sampling errors), while the clickstream data is
  discrete (by click events)
• What EDA
• Explain what analysts have been doing to analyze
  such large datasets.
• Essentially, they have been using some sort of
  statistics (descriptive, clustering, factor
  analysis, PCA Machine learning).
• Explain the pros (scalable) and cons of such
  approach (some are hypothesis instead of
  data-driven, ie. confirmatory rather than
  exploratory. As a result, it can be difficult to
  discover data pattern. No cross-checking with
  data regarding results?)
• Explain what EDA is, and how people explore data
  visually, and why is that needed in the analysis.

Venue? paper
InfoVis'06 paper rejected CHI 2007?
VSS, APGV06
Visualization for EDA
Why Visualization Explain how visualization can
help by working with established statistical
methods (EDA -gt generate hypothesis -gt checked
with statistics/further analyze data with
statistical methods -gt visually verify analysis
results -gt visually display results for
presentation/report) Explain existing EDA
visualization for large data set (TJ,
space-filling...)

• How EDA Systems
• Apply study results to an application domain
• Tools for data analysis
• Line Graph Explorer (XY data)
• Session Viewer (Web log data)
• Information retrieval/management
• music (MusicLand,
  MusicLand ??)
• email (Evita CPSC534a report)
• Previous literature search on personal
  information management for the Memoplex project

AVI06
CHI 2007

• Other related documents and sites
• Time line (Year 2)
• GPPF
• roadmap v.1

Poster InfoVis05