Transcript Breakout-Intel

Semantic Web for the Military User
Intelligence Breakout Session
Dr. Joe Rockmore/Cyladian Technology Consulting
Participants
•
•
•
•
Elaine Marsh/NRL
Frank Muller/BBN
Paul Kogut/Lockheed-Martin
Joe Rockmore/Cyladian
Charter
• How do the ideas of the semantic web specifically apply to
intelligence problems?
• What unique problems does the intelligence community
have with respect to using semantic web technology?
• How can we leverage the work being done in DAML, and
specifically the applications to intelligence, to other
efforts?
Value Propositions
• Consumers = custom products; producers = get credit for
production
• Partial automation of analysis tasks (helps info overload)
• Consolidation of data (structured and unstructured)
• Supports collaboration across orgs by common
understanding (via ontologies and inference making)
• Publish once, derived products
• Better extraction of information & “query mining”
• Feedback on missing information, including to collection
management
Semantic Web
Functional Architecture
Docs
User
interactions
Markup
Markup
Markup
{ DAML }
DBs
KB
Map
Map
Map
Analyses
Ont dev
• Browsing
• Visualization
•Q &A
• Etc.
Intelligence Ontologies
(vice C2, logistics, or others)
• .Intelligence needs to talk about what was, is, and might be
(with uncertainty), while C2 plans what to do with resources
available, logistics makes resources available, etc.
– Ontologies need to reflect differences in data and mission
•
Issues of interest to intelligence (primary)
– Money laundering, geopolitical issues, financial transactions, nonmilitary organizations, drugs, counter-terrorism, etc.
– Imagery, signals, open source, & analysis of this data
• Generally higher levels of abstraction than C2, etc.
• Source info and confidence in source important
• Temporal and spatial reasoning important
Significant Issue: Geolocation
& Temporal Representation
• Understand documents enough to know locations in a
document
– Placename, lat/lon, BE num, UTM, etc.
– Disambiguation
– Granularity issues
• Understand documents enough to know temporal aspects
in a document
– Absolute time in different granularity (date & time to milliseconds
vs. season) and representations (Julian date, DTG, etc.)
– Disambiguation
– Relative time (before, after, within, overlapping, close to, etc.)
• Coreference problems in geolocations and times
Significant Issue:
Markup Tools
• Consumer-based and producer-based markup tools needed
– Combine automated and manual markup intelligently
– Markup as part of authoring
• Culture is analysts (producers) are too busy to do any additional work,
such as markup, unless
– Its very easy to do
– There is clear value to producers (not just consumers)
– Someone measures them on the quality/quantity of markup
• Mid term: mixed initiative, where authoring and knowledge object
creation are done in parallel and with either driving the process
• A long term view: author knowledge objects from the outset; form
products from these objects, including English text documents
• Multilingual opportunities
Significant Issue:
Access to Data
• Tailored push; also pull (“My Intelink”), including changes of
sufficient magnitude
– Subscriptions and data descriptions for matching against subscriptions
may be best done using hierarchical ontologies (vice database schemata,
which are not sufficiently expressive)
• Crawlers of value, but may have access control issues (open source an
exception)
• Uncertainty of data (both by source and about source)
• Inference-based retrieval of information
• Pedigree critical to maintain (but often raises the security levels)
• Indexing of markup important for speed of access
• Timelines for intelligence information.
– Can be long, if national
– Can be short, if tactical
Significant Issue:
Collection
• Tie collection, processing, production together
• A common markup language will enhance collection, thus
optimizing use of intel resources
• Producers and consumers have different ways of looking at
the world; there is not necessarily a mapping between them
– Can consumers provide tasking to producers, via markup, of
requirements on collection?
– Info data needs from UJTL tasks or other statement of data needs
Significant issue:
Security
• Will DAML markup allow semantic understanding of
information enough to affect releasability processes?
– Can we do our collection and analysis at SCI and report at lower
levels (including collateral , coalition, and unclass)?
• [other issues]
Recommendations
• Military and intelligence users that particularly should hear
about semantic web:
–
–
–
–
–
DoD elements: DIA (esp JIVA), NSA
Agencies: NRO, NIMA, CIA
Service intel agencies: ISCOM, AFIA, ONI, MCIA
Unified commands: JIC’s and JAC’s
Standards setting and interoperability orgs
• How do organizations understand what DAML products
and approaches could help them?
– Focused TIE’s with appropriate producers and consumers around
specific value propositions
Need straightforward explanation of what
DAML is and its value added (over XML)