Knowledge Sifter: Agent-Based Search over Heterogeneous Sources using Semantic Web Services

draughtplumpInternet and Web Development

Oct 22, 2013 (3 years and 9 months ago)

79 views

1

Knowledge Sifter
:

Agent
-
Based Search over
Heterogeneous Sources using
Semantic Web Services

Faculty:

Dr. Larry Kerschberg and Dr. Daniel Menascé

Students:

Hanjo Jeong, Scott Mitchell and Ahmed Abu Jbara

Affiliates:

Drs. Riki Morikawa, Randy Howard, & Wooju Kim

E
-
Center for E
-
Business,
http://eceb.gmu.edu
/

Volgenau School of Information Technology and Engineering

George Mason University, Fairfax, Virginia


Sponsored by the NGA: National Geospatial
-
Intelligence Agency

2

Presentation Outline


Goals of the Knowledge Sifter Project;


Knowledge Sifter Architecture for semantic
querying, accessing, ranking and integrating
information from heterogeneous data sources;


Specification and design of Knowledge Sifter Meta
-
Model for storing end
-
to
-
end scenario information
in a Knowledge Repository;


Conclusions


Demonstration of Knowledge Sifter 2 Prototype.

3

Goals of Knowledge Sifter Project
-

1


To provide intelligence analysts with services to:


Specify queries related to their work tasks and


Retrieve, rank and integrate results from multiple
sources;


To use the emerging Semantic Web to create an
object
-
oriented view of people, places, things
and events by aggregating and integrating
information from multiple heterogeneous
sources.

4

Goals of Knowledge Sifter Project
-

2


To use open standards to easily incorporate new
ontologies and sources in a plug
-
and
-
play fashion:


Imagery Standards (Web Map Services and Web Feature Services,
and ISO Standards 19115 and 19139);


Semantic Web (RDF, RDFS, Web Ontology Language


OWL)


Web Services and Semantic Web Services to allow sources to be
easily discovered and incorporated into the Knowledge Sifter
architecture.


To create an agent
-
based service
-
oriented architecture that
takes user queries, enhances them semantically, submits
them for processing against multiple heterogeneous
sources, and ranks them according to the user’s
preferences and the systems similarity metrics;


5

Goals of Knowledge Sifter Project
-

3


Monitor and capture the actions and artifacts of
users, KS agents, and data sources so as to learn user
patterns, system patterns and source patterns in
order to
evolve

Knowledge Sifter over time.


To use data mining techniques on the knowledge
repository to mine patterns such as:


User preferences, contexts, and biases;


System templates for Web services choreography;


Data source QoS, availability and authoritativeness.


Monitor sources in real
-
time for QoS and adjust
query traffic using a Web services broker.

6

Knowledge Sifter Architecture


Three
-
layer
architecture:
User,
Knowledge
Management, and
Data Sources,


Autonomous agents
handle specialized
tasks,


Multiple domain
models, ontologies,
and authoritative
services;


Web services agent
handles requests to
multiple
heterogeneous
sources;


Ranking agent rates
results based on user
and system
preferences.


7

Imagery Domain Model in UML


Imagery Domain Model is the
image ontology;


An Image has several Features
such as Date and Size, with
their respective attributes.


An Image has a Source and
contains Content such as a
Person, Thing, or Place.


Classes are related by
relationships and ISA
relationships.


Classes have properties.


OWL schema of Imagery
Domain Model used by
Knowledge Sifter agents to
instantiate a query and
associated metadata.

7

8

User Layer


User Agent interacts with user to obtain
information regarding query specification;


Cooperates with Preference Agent to
provide personalized criteria for search
preferences, authoritative sites, and result
ranking evaluation rules;


Cooperates with Query Formulation Agent
to convey user preferences and the user’s
initial query.


9

Knowledge Management Layer


Query Formulation Agent consults the Ontology
Agent to enhance the query with “semantic”
concepts.


Ontology Agent uses Imagery Domain Model,
authoritative name services, and associated
ontologies to specify semantic search concepts and
coordinates for objects of interest.


Authoritative Name Services include WordNet from
Princeton University, GNIS from USGS, and GEONet
from NGA.


Query Formulation Agent receives the semantic
query and passes it to the Web Services Agent for
processing.


10

Knowledge Management Layer


Web Services Agent


Decomposes the query into subqueries and determines which Web
Services or wrapped sources should process the sub
-
queries;


Translates the subqueries into query format of local sources;


In the case of Web Services such as TerraServer, uses the SOAP
message format specified by the WSDL;


Selects the appropriate data source with semantic quality, QoS, and
availability factors;


Handles the choreography of web services execution;


Results, returned to Web Services Agent, are then sent to the Ranking
Agent.


Ranking Agent


Ranks the resulting information according to user ranking
preferences, source authoritativeness, similarity measures, etc.


Sends results to User Agent for presentation to the user.


11

Data Sources Layer


Web Services and wrappers used to link to data
sources;


Heterogeneous data sources include,


Image metadata, image archives, XML
-
repositories,
relational databases, the Web and the emerging Semantic
Web.


Quality of Service Issues


Specification of performance and availability QoS goals.


QoS negotiation protocols.


Hierarchical caching to support scalability.


12

Knowledge Sifter Meta
-
Model


Meta
-
model
describes agent
interaction, KS
artifacts, feedback
by users, etc.


Meta
-
model serves
as a schema for
capturing and
storing artifacts
such as user queries,
reformulated
queries, data sources
used, query results,
ranked results, user
feedback, etc.

13

Knowledge Sifter Meta
-
Model

14

Protégé Meta
-
Model Ontology

14

15

KS Meta
-
Model OWL Specification


OWL & RDF
Specification generated
automatically from
Protégé specification.


Meta
-
model guides the
functioning of Knowledge
Sifter and captures the
relevant data from the
actual agent
-
based
execution.


Data stored in MySQL
database according to a
relational database for
meta
-
model.


16

Conclusions


The goals of Knowledge Sifter are to provide services for analysts
to pose semantic queries to multiple heterogeneous sources
without regard to the format or location of those resources.


KS is based on open standards


Imagery, Semantic Web and Web
Services


allowing a plug
-
and
-
play semantic architecture.


KS uses authoritative name services to provide concept synonyms
(WordNet), and object location services (GNIS and GNS).


KS sources are accessed via Web service API (TerraServer) or via
wrappers.


Longer
-
term research will focus on:


Support for emergent semantics and evolution;


Collaborative filtering to inform users when others are interested in similar
concepts; and


Mechanisms by which analysts may specify hypotheses or scenarios, and
the evidence will be drawn from the multiple heterogeneous sources.

17

Knowledge Sifter Publications

http://eceb.gmu.edu/publications.html


L. Kerschberg, M. Chowdhury, A. Damiano, H. Jeong, S. Mitchell, J. Si,
and S. Smith, “Knowledge Sifter: Agent
-
Based Ontology
-
Driven Search
over Heterogeneous Databases using Semantic Web Services,” in
Semantics for a Networked World, Semantics for the Grid Databases,
LNCS 3226, vol. 276
-
293, Lecture Notes in Computer Science, M.
Bouzeghoub, C. Goble, V. Kashyap, and S. Spaccapietra, Eds., LNCS
3226 Paris, France: Springer, 2004, pp. 278
-
295.


L. Kerschberg, H. Jeong, and W. Kim, “Emergent Semantics in
Knowledge Sifter: An Evolutionary Search Agent based on Semantic Web
Services,”
Journal of Data Semantics
, Springer, 2006.


L. Kerschberg and H. Jeong, “Just
-
in
-
Time Knowledge Management,”
Keynote Talk, Third Conference on Professional Knowledge
Management, April 10
-
13, 2005, Kaiserslautern, Germany.


L. Kerschberg and H. Jeong, “Ubiquitous Data Management in
Knowledge Sifter via Data
-
DNA,” International Workshop on Ubiquitous
Data Management (UDM2005), Tokyo, Japan, April 4, 2005

17

18

Doctoral Dissertations


Dr. Mohamed N. Bennani, “Autonomic Computing through
Analytic Performance Models”, May 2006. His advisor was Dr.
Menascé.


Dr. Monchai Sopitakmol, “Experimental Study of Performance
Sensitivity of Configurable Parameters of Web
-
based Systems”
November 2004. His advisor was Dr. Menascé.


Dr. Randy Howard, “A Knowledge
-
Based Framework for
Dynamic Semantic Web Services within Virtual Organizations”
October 2004. His advisor was Dr. Kerschberg.


Dr. Riki Morikawa, “A Framework for an Analytical Knowledge
Base that Combines XML Topic Maps, Bayesian Networks, and
the Concept of Network Scenarios for Enhanced Knowledge
Sharing” July 2004. His advisor was Dr. Kerschberg.


19

Knowledge Sifter 2

Proof
-
of
-
Concept
Demonstration

http://knowledgesifter.gmu.edu/

20

KS Main Page for Query: Rushmore

21

KS User Preference Pane

22

KS Google Map for “Rushmore, Mount” in SD, US.


23

KS Google Earth for “Rushmore, Mount” in SD, US.


24

KS Google Earth Page for “Rushmore Farm” in Zambia.


25

KS Image Results for “Rushmore, Mount” in SD, US.