Semantic Web Technologies
In Practice

Bijan Parsia

Semantic Web Technologies

• "The Semantic Web"
• Languages
• RDF(S), OWL, SPARQL
• RDFa, GRDDL, etc.
• Infrastructure
• Parsers & APIs
• Reasoners and query engines
• Debuggers, model and module extractors, diffing, profiling...
• Editors, browsers, visualizers
• Applications!

About me

• Founder and Chief Scientist at Clark & Parsia, LLC
• 3 employees + founders (and growing)
• Customers include NASA, NCI, NATO, defense and telecom companies
• Custom applications, research & development, tool infrastructure
• Lecturer at University of Manchester
• Standards Vet
• OWL, WSDL, WS-Policy, SPARQL, OWL 1.1, RIF, OWL-S, SWSI, etc.
• Philosopher in Exile

A Simple Taxonomy

• Needle in the Haystack
• Known vs. Unknown Needle
• Building the Stack
• Analysis
• Terminology Engineering
• Building
• Deploying

Garlik.com 

• (Known) Needle in the Haystack
• UK Based tech startup
• “give people real power over their online data” 
• $18.5m in venture capital 
• Incorporates members from the 3Store team
• DataPatrol
• Reports on personal information online 
• Uses SPARQL to build these reports 
• Currently 57,000 users! 
• See the advert
• http://www.garlik.com/datapatroladvanced.php
• Key developer, Steve Harris, member of DAWG

Garlik: Tech details

• Reports 
• 500-2000 SPARQL queries to build a report 
• Often recursive, i.e., using prior results to find next ones 
• 8 knowledge bases of 2 billion triples each 
• Reports take 1-2 seconds to generate
• Query characteristics 
• Highly heterogenous 
• Lots of GRAPH and OPTIONAL 
• Some FILTER and ORDER BY 
• Results 
• XML Format but not the protocol (for performance)

A Building Block: JSpace 

• Front ends critical!
  
• Flexible browsing 
• http://clarkparsia.com/jspace 
• The mSpace approach developed at U. of Southampton
• “Google meets iTunes”
• Browsing rather than querying 
• Tame multi-dimensional data
• Easy perspective shifts
• Selections drive query building 
• Each column selection instantiates a variable and
  adds some conjuncts
• One can browse intermediate results

POPS (a jSpace app)

• (Unknown) Needle in the Haystack
• Expertise location service for NASA 
• Roladex culture 
• Serendipity is key 
• Hence browsing over search
• Federates 4 diverse data sources 
• 4.5M triples 
• Most queries are built by browsing 
• Fixed queries for info pane and socnet 
• Pilot for Office of the Chief Engineer
• Production will see 10,000 users
  
  

A POPS Note

• Some simple advantages of RDF(S)
• Object model helpful
• URIs as OIDs
• Attributes and Values
• Links!
• Class model useful
• If only for consistency
• E.g., faced with ad hoc inheritance models
• Standard solutions often better
• Exploit modeling affordances

BIANCA

• Business Impact Analysis for Network
  Computer Assets 
• Integrated view of applications, servers, networks,
  and changes, and their relations 
• Supports interruption analysis
• Sensitive data, so few users (~50) but high impact 
• One of the first deployed SemWeb Apps at NASA 
• Tech details 
• 100,000 triples 
• 6-8 sorts of queries 
• Classification tree, instance retrieval, graph building

BIANCA 

• Interruption analysis
• "Semantically" Shallow Analysis
• I.e., graph tracing to find dependancies
• RDF(S) level modeling
• Simply making the data accessible is important!
• Expansion paths
• "Deeper" analyses
• Interaction with security policies
• Incorporating background knowledge
• Build more complex organization model
• Build more complex system/application model

("Semantically") Deep Analysis

• XACML, WS-Policy, etc.
• Languages for expressing policy constraints
• Systems generally focus on enforcement
• Policy development and maintenance need:
• Methodologies
• Services
• Services explored by reduction to formalisms
• Such as OWL
• Also suggestive of methodological points
• Iterative refinement
• Auto-discovery of cross-cutting concerns

Terminologies

• Bio-ontologies, esp. bio-medical ontologies, long been the most significant application area
• SNOMED-CT, GO, Galen, FMA, NCI Thesaurus, etc. etc. etc.
• Some characteristics
• Very large; largeness is the watchword
• Collaboratively developed
• Long-lived
• Often complex modeling
• Though highly patterned
• Applications (and application characteristics) vary enormously
• Better infrastructure is critical!

Terminology Development

• NCI Thesaurus
• 50,000 concepts
• Team of up to 20
• Provide terminology for application builders
• Cancer Image Database
• Cancer Data Standards Repository
• Recently migrated to open source, standards based tools
• Protégé, FaCT++, and (more recently) Pellet
• Heavy investment in tool infrastructure
• Most recently explanation services and incremental reasoning
• Critical to both speeding up development and improving quality

A Key Building Block: Pellet

• Pellet is a popular, open source OWL reasoner
• First to cover entire language, and 1.1
• Includes novel services
• Debugging, incremental reasoning, conj. query
• Good middleware
• Jena, OWL API interfaces
• Command line, web service
• Bundled with many editors 
• TopBraid Composer, Protege4, Swoop
• Originally developed at the University of Maryland
• Now by C&P, actively
• Original author, Evren Sirin, directs development

Performance...

Performance is an ongoing issue!

Without a solid DL foundation, the Semantic Web would have remained largely irrelevant to health care terminology standardization... The development of OWL 1.1 eliminated one of the most significant barriers to use of OWL for SNOMED, since it permits the identification of tractable sublanguages capable of handling the size and complexity of SNOMED.

Kent Spackman, An Examination of OWL and the Requirements of a Large Health Care Terminology, OWLED 2007 

...and Expressivity

People need to be able to say what they want to say.

[A]dding property chain inclusion axioms  ... was essential. Without it, adoption of OWL by the SNOMED community would have required awkward workarounds with their attendant complications and complexities – effectively killing movement in that direction. With it, we have a clear path to using OWL 1.1 for further development and integration with other biomedical ontologies.

Kent Spackman, An Examination of OWL and the Requirements of a Large Health Care Terminology, OWLED 2007

Patient Data Management

• UK NHS has a £6.2 billion “Connecting for Health” IT program
• Key component is Care Records Service (CRS)
• “interactive patient record service accessible 24/7”
• Patient data distributed across local centers in 5 regional clusters, and a national DB
• Detailed records held by local service providers
• Diverse applications support radiology, pharmacy, etc
• Applications exchange messages containing “semantically rich clinical information”
• Summaries sent to national database
• SNOMED-CT ontology provides common vocabulary for data
  

SNOMED

• Large: 373,731 concepts & over 1 million terms
• Language used corresponds to EL++ fragment of OWL 1.1
• NHS version extended to 542,380 classes with 
• 19,828 additional named classes
• 148,821 class drug taxonomy (primitive hierarchy)
• OWL reasoner (FaCT++) classified NHS ontology 
• Able to classify whole ontology in <4 hours
• Interesting results come from 19,828 additional named classes
• 180 missing subClass relationships were found, e.g.:
• Periocular_dermatitis subClassOf Disease_of_face

SNOMED

Post Coordination

• Vocabulary is extensible at point of use: “post coordination”
• Users (e.g. clinicians) may enter novel class descriptions
• Terminology service (reasoner) used to classify descriptions
• Typical new term might be “allergy caused by almond” 
• OWL reasoner (FaCT++) used to classify new term
• Able to perform classification in <10 ms
• Classified as a kind of “nut allergy”
• Clearly of crucial importance to recognize patients with allergy caused by almond as kinds of patient with nut allergy

Online Self-Medication Advice

• Self-medication is pervasive, but can be hazardous
• 180 deaths in the USA in 2006 
• French project to provide on-line advice
• Will be made available to 20 million customers of French health insurance companies
• Patients have their own simple health care record (SEHR)
• Diagnosis system considers symptom descriptions, SEHR, Q&A and self-medication KB
• Uses OWL reasoner to advise on treatment, and check for contra-indications, side-effects, etc.
• E.g., do not take x if patient suffers from y; side-effects may include z

Online Self-Medication Advice

Online Self-Medication Advice

• Data taken from drug terminologies, e.g.:
• European Pharmaceutical Market Research Association (EphMRA)
• Anatomical Therapeutic Chemical (ATC)
• Data transformed into OWL ontology
• Reasoner used to check and enhance ontology
• OWL reasoner also used to check and enhance data
• Combined with induction and interaction with expert
• Corrected missing/incorrect information on interactions, contra-indications, allergies, side-effects, etc.
• Quality of data improved by factor of 8%

OWL Community

• OWL: Experiences and Directions (OWLED) Workshops
• Users, Implementors, Theorists working to advance the state of things
• OWLED 2007 had >90 participants, >40 papers
• OWLED 2005-2006 yielded OWL 1.1 (a W3C submission)
• OWL WG charter just went to AC!
• Task forces
• http://webont.org/owled/taskforces/
• Identify widespread needs; meet them
• Query, CNL, Annotations
• Rules, Keys, Imports, Macros, Constraints, Uncertainty, Education

New OWL WG

• Strong support for extending OWL
• E.g., Siemens, Partners Healthcare, Oracle, webMethods, SRI, (IBM!) etc.
• Balance disruption against (perceived) stagnation
• Stability is good
• But don't want to seem moribund!
• OWL has stimulated a lot of interest in the technology
• E.g., Pellet wouldn't exist, if not for OWL

(Some) Other Application Areas

• Form Management
• See the Galen Experience
• Siemens picking this up
• Should apply to other areas than healthcare!
• Configuration management
• Classic application!
• Data integration
• Conceptual Modeling
• See Renault OWLED paper

Thanks

• To Steve Harris for the Garlik.com information.
• To Chris Wroe for the CRS information and Ian Horrocks for the slides.
• To Ian Horrocks for Online Self-Med Advice slides
• To Andrew Schain and Kendall Clark for Bianca and POPS discussion.

About Clark & Parsia, LLC

DC-based R&D firm specializing in Semantic Web, web services, and advanced AI technologies for federal and enterprise customers.

http://clarkparsia.com/