INTELLIGENT CONTENT MANAGEMENT SYSTEM IST200132429 ICONS

1
INTELLIGENT CONTENT MANAGEMENT
SYSTEMIST-2001-32429 ICONS

• Prof David Bell
• Dr Hui Wang
• Yaxin Bi
• Kieran Greer
• Gongde Guo
• University of Ulster

2
Status Report July to December

• Coordination on WP2
• Task 2.1 Extracting knowledge from complex
  content objects into an ontology base with logic
  inference capabilities D07 (CIES)
• Task 2.2 Mapping UML semantics into RDF D08
  (UU)
• Task 2.3 Capturing procedural knowledge from
  process class definitions and from process
  instance execution measures D09 (Rodan)
• Task 2.4 Specification of Multi-paradigm
  intelligent KR D10 (UU, Rodan, ICS, CIES)
• Contributions
• Contribute to WP1 D16 (architecture
  specifications of D-S, TCE, Agent)
• Contribute to WP5 management report (with
  respect to D-S, TCE, Agent)
• Contribute to WP9 D30 management report

3
Ongoing tasks in the third semester

• Development of Dempster-Shafer engine
• Development of Text Categorization Engine
• Intelligent Agent Development Environment

4
Dempster-Shafer Engine

• Decision making based on incomplete or uncertain
  information and knowledge is very common in daily
  business activities
• Dempster-Shafer Theory provides a effective and
  formal way for supporting decision making based
  on incomplete information and imprecise
  knowledge, and mechanism for explicitly weighting
  conflict evidence
• It formulates decision making reasoning process
  as two major factors in terms of evidence and
  hypothesis, and bases these on a strict formal
  process to infer conclusions from the given
  uncertain evidence

5
An example of D-S

• We wish to find out if the Despatch Department
  (D) is following a statutory Purchasing
  Procedure (P) deemed to be mandatory by
  management
• Problems arise in our incomplete, imprecise
  knowledge, understanding, and control of the
  environment, and in our communications, which
  lead to uncertainty, and we use an evidential
  approach to tackle this
• We can formulate the above as follows
• three pieces of evidence testimony, observation
  and telephone call
• The 2 conclusions D follows Procedure P, D does
  not follow Procedure P

6
Text Categorization Engine

• The objective of TCE will be to develop an
  automated text categorization engine using
  machine learning methods (SVMs and KNNM) that
  will inductively learn from predefined documents
  and then create a learning model (called
  classifier). This classifier will then be used to
  assign newly arriving documents (or unclassified
  documents stored in data repository) to one or
  more preexisting categories.

7
What we have achieved so far?

• We have finished the specification design and
  technical design of TCE.
• We have designed and are developing a KNN model
  based method for TCE.
• We are developing and integrating SVM based
  method for TCE.

8
The Architecture of TCE
9
Intelligent Agent Development Environment

• JADE has been chosen as the agent platform for
  the IADE.
• JADE is an Open Source project that is FIPA
  compliant.
• FIPA (Foundation for Intelligent Physical Agents)
  produces the standards for agent development.
• Agents will interface with Web Services.
• Use the Java Web Services Development Pack
  technologies to access the Web Services. In
  particular, try JAXR, JAXM and SAAJ.
• Information passed to the agents in the form of
  XPDL.

10
Intelligent Agent Development Environment (contd)

• IADE required to search for and call Web
  Services.
• Search agent will look for Web Services, that the
  user will manually accept or reject.
• Agents will be written that can connect to a
  particular Web Service.
• Look at the possibility of using ontologies to
  allow an agent to dynamically connect to a
  particular type of Web Service. Dynamically
  construct SOAP messages. Further research
  required.
• Agent societies controlled by a master agent will
  work together to complete tasks on behalf of the
  user.

11
Task 2.2 Mapping UML semantic data model into
the Resource Description Framework

• D08 Equivalence of UML and RDF
• Yaxin BI, Prof David Bell, Dr Hui Wang, Dr Kieran
  Greer, Gongde Guo

12
Objective of mapping UML to XML (DTD, Schema, and
RDF)

• Towards specification of the UML SDM mappings
  onto the XML DTD / Schema complex content data
  model and the RDF object relationship model
• Investigate equivalence of UML semantic data
  model and the RDF content model

13
Four aspects need to be addressed

• Specify a content data model in UML which will be
  used to define the ICONS Content Base
• Investigate methodologies for translating the
  object models in UML to XML DTD / Schemas complex
  content model
• Develop domain ontologies based on the content
  model represented in RDF
• Investigate mapping rules from the UML models
  into the RDF object model

14
Roadmap
transform
Content model
Object model in UML
equivalence
Object model in XML DTD / Schema
equivalence
partially extract
Object model in RDF
15
An approach to mapping UML to DTD / Schema

• Translate relational models to object models in
  UML
• Map inheritance based on a table or multiple
  tables
• Map keys and relational joins to associations
• Map "intersection tables" to object relationships
  

16
Mapping a relational content model to an object
model in UML
17
An example of mapping UML to XML DTD and Schema
lt!ELEMENT Orders (Date, CustNum)gt lt!ATTLIST
Orders OrdersNum CDATA REQUIREDgt lt!ELEMENT Date
(PCDATA)gt lt!ELEMENT CustNum (PCDATA)gt
ltxscomplexType name"OrdersType"gt
ltxssequencegt ltxselement name"Date" /gt
ltxselement name"CustNum" /gt
lt/xssequencegt ltxsattribute
name"OrdersNum" /gt lt/xscomplexTypegt
18
Two approaches to mapping from UML to RDF

• One approach is to map from visual model (graphs)
  to representation of textual description
• The other is to map from visual model to Directed
  Labelled Graph (DLG) model

19
Mapping rules
20
Mapping rules (contd)
21
Function comparisons of some tools
22
Relationship between ontologies and content
objects
Domain Application
Ontologies Data sources
Content models in RDF
Concepts Variables
Ontologies in RDF
Terms Values
23
Construct ontologies

• Derive from the semantics encoded by RDF stored
  in the Content Base
• Extract semantics from XML/RDF repositories or
  classification servers distributed on the
  Internet to complement the domain ontologies

24
Task 2.4 Specification of the ICONS
multi-paradigm integrated knowledge schema and
query language

• D10 A Multi-Paradigm Integrated Knowledge Schema
• Kieran Greer, David Bell, Hui Wang, Yaxin Bi,
  Gongde Guo (UU)
• Witold Staniszkis, Bartosz Nowicki, Mariusz
  Momotko (Rodan)

25
Introduction

• This report provides a comprehensive overview of
  the ICONS Knowledge Schema structure, associated
  semantics and consistency assertions.
• The three principal Knowledge Schema paradigmatic
  areas are the Structural Knowledge
  representation, the Declarative Knowledge
  representation, and the Procedural Knowledge
  representation areas.
• The Knowledge Maps that are to be used as
  ontological features have been given special
  attention.

26
Introduction

• This report shows tight integration of
  knowledge-based features (Datalog and
  Dempster-Schafer) with the object-oriented schema
  definition facilities adopted as the Structural
  Knowledge paradigm
• There is also the capability to provide for
  seamless integration of external data sources
  with the ICONS Content Repository.
• The extensive use of XML enhances the potential
  inter-operability with the environment.
• Web Services technology underlying data
  integration and publication should result in a
  truly open, standards-based knowledge management
  system.

27
The Multi-Paradigm Knowledge Schema
28
Knowledge Paradigms

• The Structural Knowledge representations provide
  meta-information mechanisms for modelling content
  object class relationships, content object class
  behaviours, content object class grammars
  governing the internal object structure, and the
  object categorisation maps.
• The Declarative Knowledge representations include
  facilities for modelling domain ontologies,
  features providing for declarative extraction of
  tabular data from pre-existing relational
  databases as well as from semi-structured
  information sources, and rule-based inferential
  methods supporting the content object behaviour.

29
Knowledge Paradigms

• In order to achieve a given goal, both data and
  algorithms to process this data have to be
  applied. Structural knowledge as well as
  declarative knowledge are mainly focused on
  representation of data, its meaning and
  dependencies. Procedural knowledge complements
  this knowledge focusing on algorithms or
  procedures.
• Temporal properties of Content Objects represent
  important semantic information of a knowledge
  management application and they may constitute
  important object selection criteria. This
  temporal information should also be handled.

30
Knowledge Maps

• Knowledge maps (structural knowledge) provide
  means to categorise information objects stored in
  the content repository. The imposed
  tree-structured hierarchical categories provide a
  powerful navigation and search device for
  browsing the content repository.
• A simple definition is that a knowledge map is a
  mapping between concepts and objects.
• The knowledge maps mapping domain is basically a
  set of concepts coming from some particular
  application domain or organisational realm.

31
Knowledge Schema Consistency

• We define the Knowledge Schema integrity
  constraints from two views the intra-paradigm
  point of view and the global inter-paradigm point
  of view.
• The intra-paradigm integrity constraints deal
  with the object class association structure and
  the corresponding consistency assertions defined
  within a specific knowledge representation
  paradigm.
• We adopt a uniform scheme of integrity
  constraints presentation by providing a partial
  UML model for each area of interest supplemented
  by a table of consistency assertions specified in
  the disciplined natural language.