Semantic Technology Conference notes

Semantic Wikis, Prof. Dr. Jurgen Angele, CTO ontoprise

Twine/Radar Networks - Nova Spivack

SW Tools - Zepheira, Eric Miller

Persistent Identifiers for the "Real Web", Zepheira, David Wood, Eric Miller

Freebase - Metaweb, Jamie Taylor

11:45am 

His team responsible in seeding database.

An open, shared database of the world's knowledge

Creative Commons Attribution License
Open APIs

Community built: collective editing; collaborative semantics

Consensual reality: available data; people, places, products, ...;
Called Topics (e.g., Pontifs, Art, Airplanes, Cheese, Tropical Storms)

3.4 million topics; 750K people; 450K locations; 50K companies;
40K movies; over 1000 types and 3000 properties

Initially used some wikipedia articles as seeds of freebase db

Not a formal system (e.g., Cyc, SUMO, True Knowledge, Halo)
Not a reasoning engine

Topics in Freebase are Unique (no two Topics represent the same thing)



research.freebase.com

Federated Terminology Authoring Using Semantic MediaWiki

2:00pm 

Terminologies in Health Domain:
	SNOMED CT / IHTSDO; LOINC; ICF-10, DRGs
	ICF 9/10; CPT; HL7

Need them to 
- compare
- aggregate
- interchange
- secondary uses
- linkage to decision support services

Terminologies centrally curated and distributed: 
- quality, consistency, accuracy
  consistency (including deleted an ID and never using for
	      something else in the future)

But slow to deal with user (i.e., the terminology experts) feedback
(e.g., spelling errors, relationships between codes, descriptions)
User best source of info regarding: problems, requirements, uses.

Central assurance necessary but does not scale.
Problem: how to distribute but maintain quality, consistency, accuracy.
Solution: Used semantic media wiki
SMW: collaborative ontology engineering

What to Do with an OWL Reasoner: Introduction to Pellet - Clark & Parsia, LLC, Evren Sirin

3:15pm

OWL-DL (all) reasoner (and much of OWL2)
Open source in Java: pellet.owldl.com
Bindings to Jena, OWL-API, Protege, TopBraid Composer
Next release dual licensed

Reasoning in OWL
Given set of assertions
- check consistency (no contradictions)
- infer new conclusions

Inference:
Penguin subClassOf Bird
Pablo type Penguin
-> Pablo type Bird

Inconsistency:
Bird subClassOf FlyingThing
Penguin subClassOf Bird
Penguin disjointWith FlyingThing

Obvious here, but when thousands of users, need to automate.

Features:
- consistency; classification (subclassing between classes); 
  realization (is instance)
- conjunctive sparql-dl queries 
  (combine schema and instance query)
- datatype reasoning 
  single property - e.g, small monitor has screen <- 15in
  combination properties - e.g., widescreen has height width ration < .75
- SWRL Rules
  DL-safe rules : applies only to instances
- Explaining and debugging
  Hard to understand large complex ontologies
  Pellet can answer:
    Why is a certain subclass relation inferred?
    Why is a certain ontology inconsistent?

OwlSight - ontology browser 
(to demo pellet features, especially consistency, subclassing)
written in Google GWT

Uses of Pellet:
- data integration
  describe data sources using ontologies
  define mappings between
  use reasoning to answer queries
- healthcare and life sciences
  terminology development and axiomatization
  decision support; intelligent user interfaces; info integration
  NCI, SNOMED, GALEN, FMA, OBO, ...
- Service Oriented Architecture
  input/output types
  pre/post conditions
  languages: SA-WSDL, OWL-S
  Reasoner supports : matching requests with services
                      (semi) automated service composition
- Policy Analysis
  languages: XACML, WS-Policy
  managing is hard: detect security holes, change impact analysis
  tools focus on policy enforcement (runtime) not analysis (design-time)
  Use OWL reasoning to analyze
  Map policy language to OWL (done for XACML and WS-Policy)
  Analysis: 
     policy subsumption, redundancy, incompatibility, verification, querying

- Config management:
  find set of components that satisfy requirements and constraints
  
- Probabilistic reasoning
  Many places for uncertainty : uncertain taxonomic relationships,
                                facts can be uncertain
  social network analysis; breast cancer risk assessment
  e.g., A sameAs B with N% probability
       *most* birds fly

clarkparsia.com

Q&A: Jess is forward-chaining
     Prolog is backward-chaining
     Pellet is neither. 
        It is Semantic tableau: add negation of query and search 

Smart Browser - AdaptiveBlue, Alex Iskold

Linked Data Panel

Access Control Policies and their Use in Shared Security Services - Semantic Arts, Timothy Swanson

9:45am

Message Passing architecture
"Can we use RDF and OWL to represent these messages"?

Security Terminology
- subject, object, operation : user/agent, being requested, CRUD

Motivation
                  +->   HR/DB
Bob -> dev server -->  Security Roles/DB

but if Bob leaves company still has access to Roles/DB

Rule-based security
- Want security roles linked to HR/DB

OBAC = Ontology-Based Access Control
- Users that satisfy a set of conditions can perform operations on a system

subject -> subject
object + operation -> object
subject and object are nodes in RDF graph

Turtle notation:
[] 
 a :CheckoutRequest ;
 :requestedResource :LibraryTermsOfService

...

XACML-DL - Analyzing Web Access Control Policies Using Semantic Technologies - Clark & Parsia, Michael Smith

9:45am

Tim focused on market where ACLs have not been written.
Feasible to start with RDF.

C&P - focus on existing ACL (don't rewrite to RDF, but manage)

XACML, WS-Policy
- languages for expressing policy constraints - how to enforce at runtime
- services explored by reduction to KR formalism - design time
  e.g., OWL
  auto-discovery of cross-cutting concerns
  iterative refinement

Policy Analysis
- denotes a set of "acceptable" things by describing them

XACML
- Access Control policies for distributed resources
- Supports many features
  arbitrary attributes in policies
  express negative authorization
  conflict resolution algorithms
- profiles for common methodologies
  Hierarchical Role Based Access Control

XACML language
- Policy is set of Rules
- PolicySet can hold Policies or PolicySets
- Combine algorithms enable modular:
  Permit-overrides, deny-overrides, first-applicable, ...
- Access Requests - list of attribute/value pairs:
  subject, resource, action (operation), environment (context)

Design-time Support
- detection of security holes
- change impact analysis

Services:
- Shallow testing: test subset of possible requests
- Deep testing: all possible combinations
- Comparison : 
- Verification : policy satisfy a particular property
- Incompatibility
- Redundancy
- Querying

Primary Difference between OBAC (Tim's above)
- Decisions : R1-Target SDubClasssOf: R1-Permit
- Combining algorithms
  XACML algos translated to class descriptions
  (R1-P or R2-P and not(R3-D) SubClassOf: P
  R3-D SubClassOf: D

Challenges
- Some aspects of policy analysis stretch OWL
  Non-monotonicity, built-ins (e.g., math, XPath)
  Change analysis and querying
  production/business rule like behavior
- Goal to provide as much analysis as possible in OWL-DL 
  then use other formalism (without being ad-hoc)

Pint - XACML Policy Analysis Tool

Calais - Thomson Reuters, Thomas Tague

Sindice.com - DERI, Dr. Giovanni Tummarello

Rising Stars

4:15pm

David Scott Lewis - moderator
Barney Pell - Power Set
Alex Iskold
Nova Spivack
Ian Davis
Tom Gruber

-------------------------
Barney Pell - Power Set

Queries in natural language
-> results
-> NLP processing of search results 
-> structured data

-------------------------
Alex Iskold - AdaptiveBlue

"things in pages, links and text"

brings semantics to existing web
auto evolves with web itself
facilitates contextual browsing
focus on pragmatic simple consumer verticals

-------------------------
Nova Spivack - Radar Networks - facebook

doyop.com/twine

-------------------------
Ian Davis - Talis

heritage in large metadata systems: libraries, edu, ...

talis.com/platform

-------------------------
Tom Gruber - Stealth-Company.com

what is killer SW app? search, collective intelligence, context browsing?

You life (on-line) - intelligence at the interface
language understanding, semantic search, machine learning, integrated services


HTML/HTTP:
user role: choose your path (link)
system role: connect the dots
- URI

Portal:
choose your channels
deliver the content
- frictionless broadcasting

Search engine:
state your query
find relevant content and filter
- ??

Integil at interf
live your life
tell me what i need to know, help me solve problems, meet needs, be proactive
- personalized context-aware AI

[ Carl Thompson SW tech journalist ]

Get Carl's attention:

Nova:
Need to get past 100 billion triple barrier (need trillions/federated)
Note: porn industry is generally early adopter of tech - but not SW (yet)

Alex:
iphone + SW (go to garden, point at tree, ask "what is it?)

Tom:
Don't push technology.
What are the human needs?
e.g., location graph + social graph

Ian:
Stick in domain you understand
Don't focus on technology
Use SW under the hood
eg., travel (tripit)

Barney:
Search, advertising and publishing
Plug vertical expertise into SW platform

When will Gartner's of the world start paying attention?

Ian:
??

Tom:
Old analysts don't get it

Barney:
Gartner has been covering since 1997
But no magic quandrants

New content might be better placed in context 
(rather than on site of person who created it)

Nova:
He has talked to the analysts
but they don't think of it as A space
They want to know how it will contribute to existing categories
But does their opinion matter that much?
They are playing catchup.
Let's show great practical examples.
Forget about SW and think products

Barney:
Predication: this conference will be twice the size next year

Semantic Markup of Java Source Code - Build Software LLC, Brian D. Eubanks

SA-REST - Kno.e.sis, Karthik Gomadam, Dr. Amit

ELMO: Mapping Objects to RDF - James Leigh

map Java objects --> RDF
@rdf on interface/class; field; property

RDF --> Java objects
- rdf does contain have object "behavior"
- need to associate/find behavior to match with instance data from RDF
  to create instance

Sesame memory store
- MemoryStore - good for < 1million triples
- NativeSore
- RDBMS RDF Store
- Mulgara Store - billions

Showed how to map to/from Java, Groovy, JavaScript, Ruby

Benefits:
- people with little RDF knowledge can be persisted into RDF store.

Panel

Hendler
pollcak/oracle
christine conner/dow jones
ivan/w3c
steve hall/value
gilmore
john

WHAT/HOW
Solve or add value to existing areas (don't lead with SW).
Simplify - but stick with what you know and where your passion lies.
Tools to help spread the technology.
How to connect disparate pieces of info (corporate wikis).
How do you connect customer data that lies inside corporate silos.
Semantics will destroy value before it will add value
 e.g., just put out semantic tagged item saying "I am selling X"
       reduces value of ebay
       open data reduces value of ebay, linkedin, facebook
       but then increases value of companies that embrace new open data model
Home run: figure out how to exploit open data.
non-deterministic web is a weakness 
 (fortune 500 companies need precise info when handling individual customers)
  and a strength (trend analysis, suggestions, ...)
Steve: We would like to find and fund a semantic "SharePoint"
Security belongs with the data

TopBraid tutorial - Dean Allemang

RDF is a means for distributing/reconstituting data to/from the web
DBS is used to interpret : what is the meaning of the value in row X / column Y
XMLS is used to validate : structure
Object Models define behavior
RDFS is used for inference
OWL model is used for ...
subClassOf just means rdf:type inheritance
subPropertyOf ...