Automated Data Analysis with Knowledge Ontologies

1
Automated Data Analysis with Knowledge Ontologies

• Ed Shaya and Brian ThomasUniversity ofMaryland

2
Web Ontology Language Visual Programming
Environment OWLVIPER

• The VxOs are working toward having all of the
  distributed tabular data described semantically
  (UCDs and data models) and mappings are being
  created between different specialized
  vocabularies.
• An effort is beginning for the descriptions of
  what each software tool (service) does and the
  form of its inputs and outputs.
• If data is sufficiently self describing, the
  system can suggest appropriate operations.
  (think, .pdf suffix indicates, use acroread,
  etc).
• Now, what if you can describe, in a computer
  readable way, your goals (the imagined end-
  process data) for the information that you want?

3
Goal Driven Automation

• Paths from the goals to the existing data (even
  if still in datacenters) can be discovered by
  automated analyses of transformations. These
  paths can be presented graphically and then
  pruned and adjusted by the scientist.
• Preferred pathways can be stored (compressed)
  as a new operation and reused for future uses.
• Successful methods can become templates (or
  WebServices) for others.
• Data from repositories can be well constrained to
  the needs and fused together logically.

4
Functionality of OWLViper

• Visual Non-Programming Interface
• High level interface between scientist and
  analysis software through a canvas to find and
  analyze data.
• Query distributed data centers for derived data
  using technical terms (not datacentric), as well
  as analyze and visualize.
• User draws flow diagram executes common code
  with java wrappers, Jython or WebService on data.
• Flow diagram can be stored, named, and reused.
• Various levels of autonomy
• Lowest level manual creation of a flow diagram
• Highest level state the goal and let it rip via
  OWL knowledge base. JENA and Pellet Reasoners.
• Scientific - handle unit conversions and error
  propagation, of course.

5
Ontology (OWL)

• Functions and data are placed in class structure
  (not a strict hierarchy) plus properties or
  relationships.

Code
relation
Class1
output
input
Operation
inverse
6
OWLViperLayersThe Data Hunter.
7
(No Transcript)
8
(No Transcript)
9
Example Operation

• Select property, allowed operations appear.
• Select operation, resultant data structure
  appears.

10
OWL Flow description

• The Flow on the Work Canvas can be fully
  described by OWL. Thus it can be saved in XML
  format and can define higher-level operations.
  It can also be created from scratch using N3
• Supercluster_a name Hercules hasGalaxy
  spiralGalaxy_a.
• spiralGalaxy_a hasMeasurement dismod_a, W20.
• dismod_a hasNumericValue value_a hasUnits mag.
• value_a constraint greaterThan number 10..
• dismod2distance hasInput dismod_a hasOutput
  distance.
• Parser presumes
• Distance_a implies distance_a a phydistance.
  Etc.
• Operation output goes to new data object.

11
Software repository

• Building software repository
• Bundle compiled and wrapped software ready to run
  through the OWLViper.
• Example data.
• Ontologically classed

12
Conclusion

• We have created an application for semantically
  aware discovery and analysis of many object, many
  property scientific data.
• This infrastructure can work with almost any
  ontology from most fields of science.
• We have prototyped how OWL can describe
  scientific data, the operations on trees of
  Class/property/Class objects and therefore
  multi-pathways workflows.
• As a bonus, we developed top level Ontology for
• General science, Astronomy, Physics, Statistics,
  Instrumentation, Computation, Chemistry,
  Quantity, and Units