28
OntoPlan: Knowledge Fusion Using Semantic Web Ontologies Héctor Muñoz-Avila Jeff Heflin

OntoPlan: Knowledge Fusion Using Semantic Web Ontologies

Embed Size (px)

DESCRIPTION

OntoPlan: Knowledge Fusion Using Semantic Web Ontologies. H é ctor Mu ñ oz-Avila Jeff Heflin. Overview. Motivation Background Semantic Web Ontologies Hierarchical (HTN) Plan Representation OntoPlan Architecture for Knowledge Fusion Task-Oriented Knowledge Fusion Knowledge Filtering - PowerPoint PPT Presentation

Citation preview

OntoPlan: Knowledge Fusion Using Semantic Web Ontologies

Héctor Muñoz-Avila

Jeff Heflin

Overview

• Motivation

• Background Semantic Web Ontologies Hierarchical (HTN) Plan Representation

• OntoPlan Architecture for Knowledge Fusion Task-Oriented Knowledge Fusion Knowledge Filtering Coping with Heterogeneity Dealing with dynamic Environments

• Future Work

• Final Remarks

Motivation

• Multiple, heterogeneous data sources including various kinds of sensors and databases• Bandwidth connection to some sources may be low

• Too much information may be potentially relevant

•Which information to provide to the warfighter?

J-2UGS

Low bandwidth

Challenges

• Task-Oriented Knowledge Fusion : Gap between the information available and the information needed

• Knowledge Filtering: Large number of distributed information sources

• Heterogeneity: Information sources commit to different schemas

• Dynamic environments: Information changes rapidly

• Information costs/value trade-off: latency time versus potential benefit

Semantic Web Ontologies

• Berners-Lee, et al. (Scientific American 01) The Semantic Web is not a separate Web but an extension of the current

one, in which information is given well-defined meaning, better enabling computers and people to work in cooperation.

• Ontology a logical theory that accounts for the intended meaning of a formal

vocabulary (Guarino 98) has a formal syntax and unambiguous semantics AI algorithms can compute what logically follows

• Relevance to Web: identify context provide shared definitions eases the integration of distinct resources

OWL

• Web Ontology Language released as a W3C

recommendation in February 2004

<rdf:Description rdf:about=“”> <owl:imports resource=“www.dod.mil/weapons.owl”><rdf:Description><Tank rdf:ID=“m1a1”> <name>M1A1 Abrams</name> <topSpeed>41.5</topSpeed> <hasArmament rdf:resource=“#cannon120mm”></Tank>…

<owl:Class rdf:ID=“Tank”> <rdfs:subclassOf resource=“#Armored”></owl:Class><owl:Class rdf:ID=“Armored”/><Property ID=“topSpeed”> <domain resource=“#Tank”></Property><Property ID=“hasArmament”> <rdfs:domain rdf:resource=“#Tank”> <rdfs:range rdf:resource=“#Weapon”></Property>…

imports

Weapons Ontology

Logistics DBs

OWL Inference

Bin Laden

<owl:Property rdf:ID=“head”> <rdf:subPropertyOf rdfs:resource=“member” /></owl:Property>

<owl:Class rdf:ID=“Terrorist”> <owl:sameClassAs> <owl:Restriction> <owl:onProperty rdf:resource=“member” /> <owl:someValuesFrom rdf:resource=“TerroristOrg” /> </owl:Restriction> </owl:sameClassAs></owl:Class>

Al Qaeda TerrorOrg

Terrorist

type

head

type

• The head of an organization is also a member of it

• A member of a terror organization is a terrorist

• Therefore, the head of a terror organization is a terrorist

Main point: the various sources may be heterogeneous

Hierarchical Task Networks (HTNs):Motivation

Tactical

StrategicTheater

CINC

JCS / NCAStrategicNational

JTFOperational

• Practical: Can be used to encode information extraction strategies

• Theoretical: Strictly more expressive than action-based representation

Hierarchical Task Networks (HTNs): Example

Complex tasks are decomposed into simpler ones

Launch from Carrier Battle

Group

Security force available (F)

Transport helicoptersavailable (H)

Establish ISB within Flying

Distance

alternativeCOAs

Select Helicopter Launching Base Select Helicopter Launching Base

Select possible area (A)Transport sec. force (F,A,H)

Embark sec. force (F,H)Fly(H,A)Disembark (F,H,A)

Position security force (F,A)Transport fuel to (A)

...Helicopters have air

refuel. capability (H)

Transport helicoptersavailable (H)

Hierarchical Task Networks (HTNs) : Knowledge Artifacts

Security force available (F)

Establish Base within Flying

Distance

Transport helicoptersavailable (H)

Task:

Conditions:

Select possible area(A)

Subtasks:

Transport sec. force (F,A,H)

Position security force (F,A)

OntoPlan: Combine Hierarchical Task Networks and Ontologies

• Hierarchical task networks (HTN) can be used to represent an on-going operation at different levels of abstraction

t11 t12

t1

HTN

• Objects mentioned in the tasks (e.g., resources) are terms defined in an ontology

Ontology

commit to

• Tasks in the HTN can be accomplished by other agents and/or by gathering information from other information sources. Objects used by these agents/information sources commit to their own ontologies

t21 t22

Ontologycommit to

OntoPlan: Architecture for Knowledge Fusion

HTN S1 S2 S3

Ontologies

HTN PlanGenerator Semantic Web

Mediator

Agent Planner

KB

executed plan

task

System

Message decoder

Task-Oriented Knowledge Fusion

Task: Classify a contact

Task:

Conditions:

Subtasks:

…commits to

Ontologies

S2

commits to

Goal-Oriented Knowledge Fusion (II)

Task: Classify a contact

S2

HTN

S3

Example

Task: Classify contact OntoPlanOntoPlan

msg: contact detected

Sensor Sensor J-2

Ontology

request: activate & scan

query: previous enemy activity in the region

Message decoder

inform command staff

Example (con’t)

OntoPlanOntoPlan

command

query: forces in the area

Message decoder

Task: inform troops in area about nature of contact

query: forces in the area

msg: inform forces about contact

Knowledge Filtering By Using LCW Statements

• Use meta-level information about the information maintained by the information sources

• Local completeness: the information source knows all information about a particular query.

• Example: The US Embassy in Albonia may have complete information about the threat in that country:

LCWTF(US_Tank(t) AND in-area(t,a)).

• During HTN planning LCW information may be inferred“get all available M-113 armored vehicles available at the ISB”

Example: Local Closed-Word Information

OntoPlanOntoPlan

Area J-2

Ontology

query: previous activity in the region

Ontology

Local J-2

Ontology

Ontology

lcw(enemy activity, region)

command

lcw(own activity, region)

Semantic Web Mediator

• A knowledge fusion system for the Semantic Web contains a knowledge base with meta information

completeness information relevance information

• Selects information sources and processes the query checks its Kb to find sources that have completeness information if found - selects and queries that source if not checks its KB to find sources that have relevant information if found - selects and queries those sources

• Can perform ontology-based query translation when needed

Semantic Web Knowledge Fusion

Intel

NOAA

SW Wrapper

SW Wrapper

SW Wrapper

Intel Ont

Sensor Ont

NOAA Ont Weather Ont

Threat Ont

Location Ont

commits to

commits to

commits to

extends

Ontologies

Information Analysis

Information extraction

Monitoring

extends

extends

Dealing with Dynamic Environments

• Various sources: Data feed

New events (e.g., received data from a previously unavailable sensors)

• Is the outcome invalid?Should the agent start the whole process from the

scratch?How to “safe” some effort but still guarantee accuracy

of information extracted?

Problem: Determine Effects of Changes

Task: Classify a contact

S2

HTN

S3

inform command staff

Changed!

Changed?

? ? ?

??

?

Idea: Build Structure Maintaining Dependencies

Task: Classify a contact

HTN

inform command staff

Dependency Graph

Propagating changes

Task: Classify a contact

HTN

inform command staff

Dependency Graph

Propagation Mechanism

• Based on the ideas Redux for Constrained Decision Revision (Petrie, 1992)

• Annotates all decisions made in a dependency graph

• A 1-to-1 map can be made between HTNs and the dependency graph (Xu & Muñoz-Avila, 2004)

Task

Task

Task Task

Planned Evaluation:Empirical

• Testbed:Create several information sourcesSources commit to their own OWL ontologies Sources contain HTN knowledge artifacts (represented in

OWL) about tasks they can solved• Measures:

The time required by OntoPlan to complete tasks Size of the remote data accessed The ratio of the information gathering actions over the

total number of actions in the resulting plans

Planned Evaluation:Theoretical

• Conditions for soundness

• Conditions for completeness

• Complexity

• Expected reduction in size of the search space.

Final remarks

• We propose to build a system, OntoPlan, that exhibit the following capabilities:

Goal-Oriented Knowledge Fusion. Mechanisms for reasoning on the relationship between the information-gathering search and the information gathering tasks being solved

Heterogeneity. Allow heterogeneous data sources to commit to OWL ontologies. The content of the sources themselves will be described using OWL.

Knowledge Filtering. We also propose the use of meta-level information to control search.

Dynamic repair. Use of dependency maintenance techniques to avoid starting process from the scratch when changes occur

• We built a prototype