Transcript Ontologies for the Semantic Web:Services and Events

Ontologies for
the Semantic Web:
Services and Events
Jerry R. Hobbs
USC/ISI
Marina del Rey, CA
Outline
1. Services: DAML-S (OWL-S)
An Event Ontology for video data
2. Time: DAML-Time
Plans for a Space Ontology
3. Ideas toward an Ontology of Information
Structure
Progress toward an Ontology of Commonsense
Psychology
DAML-S Acknowledgements
Many slides stolen from David Martin,
A few slides stolen from Terry Payne and Dieter Fensel
Motivating Tasks
Automatic Web service discovery
Find an airline site that accepts Diners Club
Automatic Web service invocation
Buy coach class United flight 347 for Sept 15
Automatic Web service composition
Find the cheapest flight LAX to some Washington
airport Sept 15 or 16 and buy a ticket;
Rent a compact car for when I arrive;
Reserve a room at the Arlington Hyatt.
Automatic Web service “decomposition”
Use Amazon.com to find reference
Automatic Web service execution monitoring
Has my credit card been charged yet?
Has the ticket been mailed yet?
Why Semantic Web Services?
Web Services
Dynamic
Static
UDDI, WSDL, SOAP
Intelligent Web
Services
WWW
Semantic Web
URI, HTML, HTTP
RDF, RDF(S), OWL
Shallow
Deep
Thanks to Dieter Fensel (U. of Innsbruck)
for use of this material
Bringing the web to its full potential
DAML-S Web Services Coalition
BBN: Mark Burstein
CMU: Katia Sycara, Massimo Paolucci*
ICSI: Srini Narayanan
Maryland / College Park: Bijan Parsia
Nokia: Ora Lassila
Stanford KSL: Sheila McIlraith*
SRI: David Martin*
Southampton: Terry Payne*
USC-ISI: Jerry Hobbs
Yale: Drew McDermott
*Contributor to these slides
What is OWL-S?
• Ontology Web Language for Services
• An OWL ontology/language for (formally) describing
properties and capabilities of Web services
• An approach that draws on many sources
• Description logic
• AI planning
• Workflow
• Formal process modeling
• Agents
• Web services
http://www.daml.org/services/
Layered Approach to Language Development
OWL-S: a major application of OWL
Future versions will build upon emerging layers
(e.g. DAML-Rules)
OWL-S (Services)
DAML-??? (Rules, FOL?)
DAML+OIL  OWL (Ontology)
RDFS (RDF Schema)
RDF (Resource Description Framework)
XML (Extensible Markup Language)
OWL-S Objectives
Automation of service use by software agents
• Ideal: full-fledged use of services never before
encountered:
Discovery, selection, composition, invocation, monitoring, ..
Useful in the “real world”
Compatible with industry standards
Incremental exploitation
Enable reasoning/planning about services
e.g., On-the-fly composition
Integrated use with information resources
Ease of use; powerful tools
Upper Ontology of Services
Ontology images compliments of Terry Payne,
University of Southampton
Service Profile:
“What does it do?”
High-level characterization/summary of a service
Used for
• Populating service registries
• A service can have many profiles
• Automated service discovery
• Service selection (matchmaking)
• One can derive:
• Service advertisements
• Service requests
Service Profile
Non Functional Properties
Functionality
Description
Service Profile: Functionality Description
•
Functional Specification of what the
service provides in terms of
parameters, subclassed as:
– preconditions
– inputs
– outputs
– effects
•
Summarizes the top-level Process
Service Profile:
NonFunctional Properties
•
Provides supporting information about the service.
Service Profile:
NonFunctional Properties
• These include
–
–
–
–
–
–
serviceName
textDescription
qualityRating
serviceParameter
serviceCategory
contactInformation
Service Profile:
NonFunctional Properties - Actor
Service Profile: Styles of use
• Class-hierarchical yellow pages
–
–
–
–
Implicit capability characterization
Arrangement of attributes on class hierarchy
Can use multiple inheritance
Relies primarily on “non-functional” properties
• Process summaries for planning purposes
–
–
–
–
More explicit
Inputs, outputs, preconditions, effects
Less reliance on formal hierarchical organization
Summarizes process model specs
– Relies primarily on functional description
Exploiting Profile Hierarchies
Tie in with
UDDI, UNSPSC, …
DL Basis for
matchmaking
Multiple profiles;
multiple taxonomies
Upper Ontology of Services
Service
Model
Process Model: “How does it work?”
“How does it work?”
Process
– Interpretable description of service provider’s behavior
– Tells service user how and when to interact (read/write messages)
& Process control
– Ontology of process state; supports status queries
– (stubbed out at present)
• Used for:
– Service invocation, planning/(de)composition, interoperation,
monitoring
• All processes have
– Inputs, outputs, preconditions and effects
– Function/dataflow metaphor; action/process metaphor
• Composite processes
– Control flow
– Data flow
Service Model / Process Model
Function/Dataflow Metaphor
Input:
• customer name
• origin
• destination
• weight
• pickup date
• ...
Output:
• confirmation no.
• ...
Acme Book
Truck Shipment
truck available
+
?
valid credit card
Y
N
• failure notification
•…
AI-inspired Action/Process Metaphor
Output:
• confirmation no.
• ...
Effect:
• goods at location
if successful
• credit card debited...
Input:
• customer name
• origin
• destination
• pickup date
• ...
Preconditions:
• knowledge of
the input
• ...
Acme Book
Truck Shipment
truck available
+
?
valid credit card
Output:
Effect:
Y
N
• failure notification
•…
<no effect>
Composite Process
Input &
Preconditions
•
•
•
•
•
•
•
•
•
•
Output &
Effects
AcmeTruckShpng
•
•
•
•
•
•
•
•
• confirmation no.
• ...
•
•
•
•
•
•
•
•
•
•
•
• customer name
• location
• car type
• dates
• credit card no.
• ...
www.acmecar.com
?
book car service
• failure notification
•…
• confirmation no.
• ...
• confirmation no.
• ...
• customer name
• flight numbers
• dates
• credit card no.
•
• ...
• confirmation no.
• dates
• room type
• credit card no.
• ...
www.acmehotel.com
book hotel service
?
www.acmeair.com
book flight service
?
• failure notification
•…
• failure notification
• errror information
•…
?
•
•
•
•
•
•
•
•
Atomic Process Example
<!– Atomic Process Definition - GetDesiredFlightDetails -->
<rdfs:Class rdf:ID="GetDesiredFlightDetails">
<rdfs:subClassOf rdf:resource="http://www.daml.org/Process#AtomicProcess" />
</rdfs:Class>
<!– (sample) Inputs used by atomic process GetDesiredFlightDetails -->
<rdf:Property rdf:ID="departureAirport_In">
<rdfs:subPropertyOf rdf:resource="http://www.daml.org/Process#input" />
<rdfs:domain rdf:resource="#GetDesiredFlightDetails" />
<rdfs:range rdf:resource="http://www.daml.ri.cmu.edu/ont/
DAML-S/concepts.daml#Airport" />
</rdf:Property>
<rdf:Property rdf:ID="outboundDate_In">
<rdfs:subPropertyOf rdf:resource="http://www.daml.org/Process#input" />
<rdfs:domain rdf:resource="#GetDesiredFlightDetails" />
<rdfs:range rdf:resource="http://www.daml.ri.cmu.edu/ont/
DAML-S/concepts.daml#FlightDate" />
</rdf:Property>
Airport
Flight Date
departureAirport_In
GetDesired
Flight Details
outboundDate_In
Composite Process Example
<rdfs:Class rdf:ID="BookFlight">
<rdfs:subClassOf rdf:resource="#CompositeProcess" />
<rdfs:subClassOf rdf:resource="http://www.daml.org/Process#Sequence" />
<daml:subClassOf>
<daml:Restriction>
<daml:onProperty rdf:resource="http://www.daml.org/Process#components" />
<daml:toClass>
<daml:subClassOf>
<daml:unionOf rdf:parseType="daml:collection">
<rdfs:Class rdfs:about="#GetFlightDetails" />
<rdfs:Class rdfs:about="#GetContactDetails" />
<rdfs:Class rdfs:about="#ReserveFlight" />
<rdfs:Class rdfs:about="#ConfirmReservation" />
</daml:unionOf>
Composite Process
</daml:subClassOf>
</daml:toClass>
BookFlight
</daml:Restriction>
</daml:subClassOf>
</rdfs:Class>
Get Flight
Details
Get Contact
Details
Sequence
Reserve Flight
Sequence
Sequence
Confirm
Reservation
Simple and Composite Processes
AcmeTruckShpng
ExpandedAcmeTruckShpng
Confirm Shipping Region
N
Acme
Truck
Shipping
expands
truck available
+
valid credit card
Y
Get Quote
Service
Get Shipping Dates
Book Truck Shipment
Upper Ontology of Services
Service Grounding:
“How to access it”
• Implementation-specific
• Message formatting, transport
mechanisms, protocols, serializations of
types
• Service Model + Grounding give everything
needed for using the service
• Builds upon WSDL
OWL-S / WSDL Grounding
• Web Services Description Language
– Authored by IBM, Ariba, Microsoft
– Focus of W3C Web Services Description WG
– Commercial momentum
– Specifies message syntax accepted/generated
by communication ports
– Bindings to popular message/transport
standards (SOAP, HTTP, MIME)
– Abstract “types”; extensibility elements
• Complementary with OWL-S
OWL-S / WSDL Grounding
OWL-S
Process Model
Atomic Process
Resources/Concepts
Inputs / Outputs
Message
Operation
Binding to SOAP, HTTP, etc.
WSDL
OWL-S / WSDL Grounding (cont’d)
OWL-S / WSDL Grounding (cont’d)
WSDL
Document
daml-property
inputX
outputY
daml-property
owl-s-process
Atomic
Process
<message …>
<part …>
<message …>
<part …>
<operation …>
<binding …>
Review: Upper Ontology of Services
Path of Evolution
DAML Coalition formed (February 2001)
Release 0.5 (May 2001)
Initial Profile & Process ontologies
Release 0.6 (December 2001)
Refinements to Profile & Process; Resources ontology
Release 0.7 (October 2002)
Initial DAML-S/WSDL Grounding;
Profile, Process Model refinements; more complete examples
Release 0.9 (May 2003)
DAML-S  OWL-S
Grounding: greater generality, flexibility
Initial work on expressing conditions, security
More community support (contributions pages)
Towards 1.0
Expressiveness issues; exceptions, lifecycle; process issues
Event Ontology for Video Data
(work with Ram Nevatia, USC)
Some concepts
Single-thread composite events
Multi-thread composite events
Granularity and primitive events
The functionality of events
An ontology of mobile objects
Motivation
In a convenience store surveillance videotape,
how do you tell the difference between
a legitimate purchase
a holdup
shoplifting
A language of describing the structure of
event in video data for recognition and
retrieval
Role of Ontology
Abstract Theory
Fixes meaning of
terms in common
vocabulary
Common Vocabulary
Application1
Application2
Mark-Up Language
How to Build an Ontology
What types of objects are there?
What are their attributes or properties?
What relations can obtain between various types
of objects?
What possible changes are there in properties
and relations? = Events
Reification of Events
p(x)
The state or event of p being
true of or happening to x
Various notation devices possible.
If events are your main topic, your logic
should have variables that can refer to them.
Some Terminology
Properties, Attributes, Relationships
An event is a change of state (e.g., location).
change(p(x),q(x)), change(at(x,y),at(x,z))
An agent, state or event can cause a state or event.
[change(p1(x),q1(x))] -cause-> [change(p2(y),q2(y))]
An agent is an entity that can initiate a causal chain.
a -cause-> [change(p(x),q(x))]
An action is an event with an agent.
a -cause-> [change(p(x),q(x))]
Agent
Instrument
John -cause-> change(at(H,A),at(H,W))
-cause-> change(intact(W),broken(W))
Theme/Object
Some Terminology
A process is a complex of tightly coupled events.
e1
e2
e3
e4
e5
An activity is a set of loosely coupled events and/or
processes.
{[change(p(x),q(x))], ....}
An event is a limiting case of a process; a process
is an event at a finer granularity.
Single-Thread Event Composition
for Single Agents/Objects
Sequence: open door; go thru; ...
Iteration: walking: {move left leg; move right leg}*
Alternation: take part; [toss into bin A | toss into bin B]
Conditionals: take part; square? ==> toss into bin A
round? ==> toss into bin B
Hierarchical Composition
Interruption and Resumption
Similar to DAML-S Process Model
Multi-thread Events
Each agent/objects executes its single-thread process.
Sometimes agents/objects interact (physically).
Some agents can observe other agents’/objects’
actions and events, and modify their behavior
accordingly.
Some agents can perform actions in order to be
observed: communication.
.
.
.
.
.
.
.
.
.
.
Multi-thread Events:
Example
Blender:
Cook:
Pot on
Burner:
off &
empty
off &
filled
turns on
burner
off
fills
blender
on
on
blending
turns on
blender
increasing temperature
boiling
Temporal Constraints
Many processes have temporal constraints among
component events.
Topological relations: begin, inside, end, before,
interval relations.
part appears on conveyor belt; pick it up
Duration: rhythm -- two agents keep iterations at same duration
Clock and calendar: process begins at 9:00 am
DAML-Time Ontology for relating events and describing their
temporal properties
Granularity
Coarse-grained view:
PERSON
CROWD
Fine-grained view:
person1
head
arm
arm
torso
person4
person2
person3
person6
leg
leg
person5
Granularity
COARSE: Complex object as single entity.
FINE: Complex object as multiple independent,
coordinated agents.
Granularity
COARSE: Complex object as single entity.
INTERMEDIATE: Complex object as single
entity with parts
FINE: Complex object as multiple independent,
coordinated agents.
Primitive Actions
Granularity-dependent, domain-dependent.
Some primitive actions in visual applications:
Cars: move w speed in direction, change
People in indoor/outdoor surveillance:
tens, not hundreds
walk, look at objects, pick up and carry objects,
hand objects to other persons, ...
People in meetings:
Gaze direction,body position,
Hand gestures: deictic, iconic, “metaphorical”, ...
The Functionality of Events
Structure: How events decompose into subevents.
Function: How events, undecomposed, relate to
their environment.
Communication
Physical
Intentional
drinking cup of
coffee
Nonintentional
fall off chair
Physical: personal
pick something up
utterance
yawn?
interpersonal
hand it to someone
Communication: medium: spoken in person, phone, whiteboard,...
audience: one’s self, one other, whole group,...
A Simple Ontology of Mobile Objects
Change + ....
MobileObject: velocity, direction
MobileObject-Object: distance-from, near
Container: portal (open/closed)
MobileObject-Container: in/out
MobileObject-PortableObject: hold
MobileObject-MobileObject: relations are all derived
from above
Derived notions: start, speed-up, turn, approach,
enter, exit, pick-up, carry, transfer, exchange
Example: Tailgating
Field of vision has regions that are restricted, unrestricted
Persons (MobileObject): entry to restricted area can be
authorized/unauthorized
Tailgate: Person1 moves thru portal into restricted area
without authorization by staying near Person2 who
causes portal to be open by establishing authorization
Functional inference: intrusion
Summary
Ontology of events, processes, and services
is useful in Semantic Web for
recognition/discovery/analysis
retrieval
composition of novel processes
decomposition into parts