Feb5-DSC-Ontology-Wo.. - Buffalo Ontology Site

Report
11:00 Self-Introductions
11:15 Report on ontology-based data integration work in DCGS-A
--- Goals and methodology
--- Practical experience and results so far
--- Risks and problems
12:15 Lunch (Brown Bag Lunch?)
13:00 Discussion of NSA work and goals (to be expanded)
--- ontology alignment / management
--- OMaaS (Object Management as a Service)
14:00-15:30 Discussion of how collaborative ontology effort (e.g.
between I2WD and NSA) will work in practice
--- how to ensure consistency (architectural and content)
--- how to address governance
--- business development and work flow
DSC Ontology Work
1. Goals and methodology
2. Practical experience and results so far
3. Challenges
02/05/13
Goal: To realize Horizontal Integration(HI)
of intelligence data
HI =Def. the ability to exploit multiple data sources
as if they are one
 Problem: the data coming onstream are out of our
control
 Any strategy for HI must be agile in the sense that
it can be quickly extended to new zones of
emerging data according to need
 Ontology can provide the needed agility and
(incremental approach to) comprehensiveness
The Business Case
• Huge resources are wasted as multiple different
agencies create lexicons, glossaries, data models,
messaging and exchange standards with the same or
closely overlapping coverage.
• Additional resources are wasted in creating mappings
between these artifacts, and in maintaining them in
light of new needs and challenges. These mappings
always fail.
• A sensible solution must incorporate an evolutionary
process which will ensure that artifacts used to manage
data converge over time.
Methodology: Create an agile strategy for building
ontologies within a Shared Semantic Resource
(SSR) and apply and extend these ontologies to
annotate new source data as they come onstream
⁻ Strategy pioneered in biomedical and other scientific
fields: leaves data as they are, and incrementally tags
data sources with terms from a growing, consistent,
non-redundant set of ontologies
⁻ Problem: Given the immense and growing variety of
data sources, the development methodology must be
applied by multiple different groups
⁻ How to manage collaboration? This is what this meeting
is about
Current State
2010-2011
– Architectural implementation (DRIF) to create the Dataspace (a cloud of
intelligence data) = lossless representation of sources with their native
semantics
– Initiated Semantic Enhancement (SE) = using ontologies to annotate these
native semantics
– Demonstration of use of SE to index and query the content of the Dataspace
2012
– Created methodology and architecture for ontology development
– Initiated Shared Semantic Resource (SSR); created suite of prototype ontologies
enabling SE also outside the Dataspace
– CUBRC: demonstrated ability to leverage SE for analytics
– Priming potential users of SE (in DoD CIO, NGA, JIEDDO, TRADOC …)
2013– Milportal
– Event Reporting Application
– Initial negotiations with other agencies and groups
Ontologies We Have (Feb. 2013)
• Physical Artifact, including Infrastructure, Facility,
Vehicle, Weapon
• Information Artifact, including Report, Image, Map
• Event, including Military, Criminal, Economic, Political,
Religious, Social
•
•
•
•
Human Physical Characteristics
Agent, including Person, Organization, Social Network
Geospatial
Time
MilPortal
http://milportal.ncor.buffalo.edu
Next step
slide from Margaret Storey
Work plans 1: For DSC Cloud (on-going)
• Perform needs analysis: Review DCGS-A Logical Data Model and
schemas of other DSCG-A data sets; in each case, examine content
and establish what terms are needed to ensure sufficient coverage
for SE;
• Where these terms already exist within the SSR, check that
definitions exist and that these definitions are adequate
• Where the terms do not exist within the SSR, create new terms (or
new ontologies), with appropriate definitions as necessary to fill
gaps.
• Use the terms in 2. and 3. to annotate the corresponding entries in
the data models to effect horizontal integration;
• With each new expansion in scope of DSGS-A data sets, iterate the
above as needed.
• In addition, we are engaging in documentation of the methodology
as here described, and in dissemination and training.
Work plans 2: For interagency
collaboration
Step 1: Initiating interagency collaboration in the service of horizontal
integration of intelligence data
Identify candidate teams / agencies
Establish collaboration with one or more specific teams
– Formulate and ratify agreements as appropriate
– Create work plan and identify funding needs
– Perform risk assessment
Step 2: Establishment of the inter-agency ontology development process
Examples of types of work to be performed would include:
• Create governance infrastructure
• Establish needed technological support
• Implement workflow
• Conduct training in methodology where needed
• Explore opportunities for inter-agency HI of data
• Begin application to relevant agency data models of the SE strategy
• Dissemination of results in a form which will allow improved systems to
perform enhanced analytics exploiting semantic interoperability.
The SSR methodology and governance is
neutral as to how SSR-annotations are used
• Currently SE of DSC enhancement only
integration through improved indexing/search
capability
• On CUBRC project, we have much more, including
ontology-based reasoning.
• In the future we will have for DSC enhancement
applied
– to multiple models such as LDM
– to unstructured text
• see the various methodology documents
provided so far
Event Reporting Application (EvR)
• Aggregates content from various report
generating applications (CPOF, TIGR, TIGR with
MAPHT extension)
• The underlying data model contains nearly
500 terms (e.g. ReportName, EventName,
DeclassificationEvent)
• The semantics of the data model seems
similar to that of a relational model
EvR Data Model as Relational Database
• Hierarchies of events and units are referenced by EventType and UnitEchelon
columns, but this alone provides no capability for traversing these hierarchies
• A report is related to both a report name and a report unit by the inclusion of
appropriately named columns, but what is the difference between these
relationships?
• Our approach shows how to express the difference, e.g. between report-toevent and event-to unit relationships
Report
Event
UnitInformation
ReportId
ID
UIC
ReportName
UnitIdentificationCode
UnitName
ReportUnit
Location
UnitAffiliation
ContainedEventIds
EventType
UnitEchelon
…
…
…
Enhancing the vocabulary of the
EvR Data Model
• Semantic enhancement (SE) amounts to associating a
database field to a whole knowledge system enabling
machines to process data …
– “vertically” e.g. by traversing the echelon command
hierarchy and
– “horizontally” e.g. by following specified relations between
units and events.
• SE separates semantics from structure, reducing
maintenance costs as source databases no longer have
to be modified each time we improve our
understanding of reality
• The ontology tags move with the data …
Enhancing the vocabulary of the
EvR Data Model
Progress to Date
• Two techniques of SE are available
– Partial enhancement (now being used in the DSC)
associating ontology label terms with EvR terms
– Full enhancement (not yet implemented in the DSC)
aligning terms from the EvR to assertions using terms and
relations from SSR ontologies
• At present the current ontologies…
– provide 70% coverage for partial enhancement
– provide 38% coverage for full enhancement
– plan for extending the ontologies to raise the coverage for
the EvR to 90% and 60% respectively by end of February
Enhancing the vocabulary of the
EvR Data Model
Event Reporting
Term
Partial Enhancement
via Ontology Label
Full Enhancement via
Ontology Assertion(s)
Enhanced EvR Data as a Graph
Event
ID
Event
Type
Unit
Identification
Code
Report
ID
Event
Report
Unit
Report
Name
Act of
Reporting
Command
Unit
Unit
Name
agent_in
Report
Unit
Service
shows how full annotation provides the semantics missing from the
EvR relational model described above
Event
ID
Event
Type
Unit
Identification
Code
Report
ID
Event
Report
Unit
Report
Name
Act of
Reporting
Command
Unit
Unit
Name
agent_in
Report
Unit
relationships between
• event type and echelon
• report name and reporting unit
are made distinct and machine processable
Service
Challenges
• Challenges to Horizontal Integration in general
– Too many lexicons
– The scope of the domain: signal, sensor, image, …
intelligence about the whole world
• Our solution
– Incremental extraction and sanitization, and creation of
content
– Distributed collaborative development
– Strong methodology
• Challenges for our solution
– Governance and management of ontology development to
ensure consistent evolution
– Lack of expertise
• For ontology development and management
• For annotation
– Success will breed failure
We have these precursor ontologies
• The prototype ontologies in the existing SE
(Malyuta-Salmen) helped indexing
• The ontologies we now have are much better
and more than these prototypes
• How should we implement them re Event
Reporting Ontology, global graph, etc. …?
• Next steps with I2WD
Our ideas are being heard
• 2 classes of collaborator: observers and
partners
• companies such as DataTactics are realizing
that in case of success this provides huge
potential business benefits.
• (DT invited our team to talk at the Ontology
Summit they have called on Feb. 12, devoted
to the development of shared semantics)
Peter Morosoff and Bill Mandrick
• How to create a strategy?

similar documents