OnToKnowledge

1

• On-To-Knowledge
• IST-1999-10132
• Content-driven Knowledge Management through
  Evolving Ontologies
• Rob Engels, Dieter Fensel, Frank van
  Harmelen, Victor Iosif, Arjohn Kampman, Uwe
  Krohn, Ulrich Reimer, Rudi Studer and York Sure
• www.ontoknowledge.org

2
Contents

• The overall goals
• The overall architecture and language
• Ontology building and instantiation
• Storing and manipulating meta information
• Querying the semantic web
• Case Studies
• Conclusions

3
1. The overall goals

• The competitiveness of companies active in areas
  with high change rate depends heavily on how they
  maintain and access their knowledge.
• Large Companies have intranets with several
  million pages. Finding, creating and maintaining
  information is a rather hard problem in this
  weakly structured representation media.
• Knowledge Management is about acquiring,
  maintaining, and accessing knowledge of an
  organization.

4
The overall goals

• With the large number of on-line documents
  several document management systems arose.
  However these systems have severe weaknesses
• Word matching as search method.
• Information retrieval instead of query answering.
• Document exchange between departments is only
  possible with severe effort.
• Different views on documents are not supported.
• Information maintenance is not supported.

5
The overall goals

• Ontologies will allow structural and semantic
  definitions of documents providing completely new
  possibilities compared with existing document
  management systems
• Intelligent search instead of keyword matching.
• Query answering instead of information retrieval.
• Document exchange between departments via
  transformation operators.
• Definition of views on documents.
• Support of information maintenance.

6
The overall goals

• The goal of the On-To-Knowledge project is to
  support efficient and effective knowledge
  management. It focuses on acquiring,
  representing, and accessing weakly-structured
  on-line information sources

• Acquiring Text mining and extraction techniques
  are applied to extract semantic information from
  textual information.
• Representing XML, RDF, and OIL are used for
  describing syntax and semantics of
  semi-structured information sources.
• Accessing Novel semantic web search technology
  and knowledge sharing facilities.

7
2. The overall architecture and language

• The On-To-Knowledge tool suite consists of
• an Ontology-based knowledge sharing facility
• an Ontology-based presentation platform
• an Ontology-based search engine
• an Ontology editor and semi-automatic Ontology
  construction tools
• inference engines and query engine for meta data
  and schema information
• persistent storage of Ontologies and meta data
  and
• extraction tools for meta data.

8
The overall architecture and language

• Open architecture, maximal reliance on existing
  standards
• XML, RDF, HTTP, SOAP, JDBC, RQL,
• Client-server approach
• allows tools to be used over the Internet
• requires minimal installation of tools locally

• All tools use DAMLOIL as ontology language
• OIL Core is the minimum requirement
• Tool scalability is targeted at supporting
• O(103) classes
• O(105) data statements

9
The overall architecture and language
10
The overall architecture and language OIL
(Ontology Inference Layer)

• RDF provides a simple data model for representing
  formal semantics of information, i.e.
  meta-information.

• RDF Schema defines a simple ontology modeling
  language on top of RDF that can be used to define
  vcabulary and structure of meta information.
• OIL adds a simple Description Logic to RDF
  Schema It allows to define axioms that logically
  describe classes, properties, and their
  hierarchies.
• Currently, a sub dialect of OIL called DAMLOIL
  is the starting point of a web ontology language
  standardization group of the W3C which should
  start soon.

11
The overall architecture and language OIL
(Ontology Inference Layer)
CreditsThanks to Ian Horrocks from Manchester!
12
The overall architecture and language OIL
(Ontology Inference Layer)

• OIL provides a layered architecture that offers
  different layers of complexity

13
3. Ontology building and instantiation

• OntoEdit Manual building of Ontologies.
• OntoExtract Semi-automatic Ontology construction
  from natural language sources.
• OntoWrapper Semi-automatic Ontology construction
  from semi-structured and structured information
  sources.

14
Ontology building and instantiation OntoEdit

• OntoEdit is a graphical Ontology Engineering
  Environment
• Helps in creating, modifying and browsing of
  ontologies.
• It is flexible and expandable through a plug-in
  framework.

15
Ontology building and instantiation OntoEdit
16
Ontology building and instantiation CORPORUM

• Structured documents Ontowrapper and
  screen-scraping extract information from places
  on specific sites (e.g. names, email addresses,
  telephone numbers).
• Unstructured documents OntoExtract extracts
  initial ontologies from natural language on web
  pages. OntoExtract is able to
• provide initial ontologies,
• refine existing ontologies,
• find relations between key terms in documents,
• find instances of concepts within document.

17
Ontology building and instantiation CORPORUM

• CORPORUMs linguistic functionality is based on a
  tokenizer, a morphologic component, and a
  relation determining engine.
• This allows CORPORUM to extract concepts from
  texts that are more then just words, concepts can
  also be generated by the engine.
• Relations between such concepts are defined (e.g.
  subClassOf relations, or InstanceOf relations).
• Through semantic analysis of a domain, the tool
  can automatically generate thesauri of words
  within a domain.
• Visualisation of such semantic structures can
  than be used for navigation and browsing through
  document sets.

18
Corporum-OntoExtract

• OntoExtract allows for analysis of natural
  language.
• The component exports its Central Concept Area in
  the form of a light-weight ontology (syntax
  DAMLOIL).
• It finds classes, sub-class relationships and
  instances.
• Finally there are cross-taxonomic relations
  describing relations between concepts that are
  often not easily recovered from standard
  ontologies.

19
Corporum-OntoExtract

• How does OntoExtract currently work
• parses, tokenizes and analyses text,
• generates nodes and relations between them,
• enhances specific aspects of the discovered
  knowledge item using a background repository
  (containing general knowledge of the world,
  represented in Sesame),
• and the final analysis results are submitted to
  the RDFS server Sesame.

20
Sesame domain knowledge
Sesame background knowledge
21
(No Transcript)
22
4. Storing and manipulating meta information

• Sesame A repository and querying facility for
  RDF Schema.
• Functionality
• Persistent storage of RDF data and RDF Schema.
• Query engine for RDF Query Language (RQL).
• Data upload and download in RDF format.
• Communication over HTTP.

Worlds first!

• Features
• Designed for scalability.
• Independent of repository types (databases,
  files, in-memory data structures, ...).
• Modular design allows for other functional
  modules.
• Architecture allows support for other
  communication protocols.

23
Sesame architecture
HTTP Handler
??? Handler
Protocol handlers
Routes requests to modules
Request Router
Functional modules
RDF Admin
RQL Engine
??? Module
Provides database independence
Repository Abstraction Layer
Repository
Persistent storage
24
Sesame RQL query engine

• RQL
• tailored to the RDF graph model
• currently the only query language for RDF Schema
  semantics

• based on OQL with features like
• Set of core queries (Class, Property, subClassOf,
  )
• Filters (select-from-where)
• Boolean expressions
• Functional composition of queries
• supports path expressions
• For navigating the RDF graph model
• Allowing mixed RDF data and schema queries

25
Sesame RQL query examples

• Basic queries

26
5. Querying the Semantic Web

• OTK provides as user access
• Querying the Semantic Web
• RDFferret
• An Ontology-based presentation platform
  Spectacle
• An Ontology-based knowledge sharing facility
  OntoShare

27
Querying the Semantic Web RDFferret

• Impractical to create RDF annotations that
  exhaustively cover the content of a given
  document
• RDF searches might produce low recall
• RDF searches produce high precision
• Search both RDF annotations and text content
• Use well proven IR techniques (ranking, stemming,
  ...)

Low threshold
28
Querying the Semantic Web RDFferret
29
Ontology-based presentation Spectacle

• Spectacle personalized information disclosure
• Personalization is ontology-based
• Spectacle can personalize
• the content itself (WHAT)
• the content presentation (HOW)
• the content organisation/navigation (WHERE)
• Example personalizations based on
• Experience beginner vs expert user
• Role maintainer vs end-user
• Task learning vs problem solving
• Etc.

30
Ontology-based presentation Spectacle
information sources
presentation data
profiles navigation layout
RDF
classification
classified data
DBMS
docs
presentation
ontology-based information presentation
31
Proactively sharing informationOntoShare

• Sharing information through an organisation is a
  key knowledge management issue
• OntoShare supports and encourages information
  sharing
• User requests OntoShare to share some information
• On sharing, page is assigned to an ontological
  category (class) and matched against each users
  ontology-based profile.
• OntoShare automatically extracts keywords
  summaries from the information and suggests
  changes to user profile based on user activity
• OntoShare proactively emails selected users when
  information of interest is shared

32
6. Case Studies

• Large Intranets Swiss Life
• Customer Relationship Management BT
• Virtual Enterprise Enersearch
• A general Methodology AIFB

33
Large Intranets Swiss Life

• Swiss Life is a large insurance company with a
  huge intranet and other distributed information
  sources.
• Efficient knowledge management for this
  information is of high strategic importance.
• OTK technology is applied in two case studies
  with Swiss Life
• Searching a Large Document on IAS (International
  Accounting Standard).
• Skills Management (SkiM).

34
Searching a Large Document on IAS (International
Accounting Standard)

• Goals
• Fast and reliable access to relevant passages
  ofa large document on IAS

• Approach
• Automatic generation of a light-weight ontology
  with weighted semantic associations between
  concepts
• Use of that ontology to support query
  reformulationby adding relevant ontology terms

35
Skills Management (SkiM)

• Goals
• Access the knowledge and skills of employees
• Identify, manage, use and advance skills

• Approach
• Use of manually built ontologies
• - to describe skills, job functions, education
  in a controlled vocabulary
• - to generate annotated homepages from the
  skills descriptions
• Exploit ontologies for a more specific search on
  homepages for people with certain skills

36
Customer Relationship Management BT

• Disseminating Customer Handling Rules
• Offer a cost-effective channel for the
  dissemination of all sorts of rules and
  instructions.
• Health Safety.
• Sales scripts for Telesales Representatives.
• Timely information on new products and services.
• Dissemination Best Practice
• Promote behaviours that are acceptably consistent
  across call centres.
• Help managers to become more aware of Best
  Practice resources on the BT Intranet.
• Help to build communities of best practice.

37
Customer Relationship Management BT

• Staying Alert-Interest Profiles
• Learn a users interests and preferences
  autonomously (with minimal feedback from the
  user).
• Adapt to changing needs of the user over time.
• Where possible user profiles should be acquired
  automatically with the users role being one of
  review to correct/refine their profile.

38
Virtual Enterprise EnerSearch

• Goals
• Improve usefulness of EnS website by semantic
  methods
• Especially for the shareholder representatives in
  virtual organizations

• Approach
• Ontology development
• Manual (OntoEdit/AIFB)
• Automatic (OntoExtract/CognIT)
• Information modes (i) key-word search (ii)
  semantic ontology-based search (RDF-ferret) (iii)
  browsing with knowledge visualization (spectacle)
• Evaluation by end user studies (i) pre-trial
  interviews (ii) end user test (iii) post trial
  studies

39
KMMethodology
Maintenance Evolution
Ontology Kickoff
Feasibility study
Refinement
Evaluation

• Check requirements
• Test in target application
• Analyze usage patterns
• Deployment

• Requirement specification
• Analyze knowledge sources
• Develop baseline ontology

• Knowledge elicitation with domain experts
• Develop and refine target ontology

• Manage organizational maintenance process (Who is
  responsible? How is it done?)

• Identify people
• Focus domain
• Select tools from OTK tool suite
• GO / No GO decision

40
7. Conclusions

• The semantic web is based on machine-processable
  semantics of data.
• It will significantly change our information
  access based on a higher level of service
  provided by computers.
• It is based on new web languages such as XML,
  RDF, and OIL, and tools that make use of these
  languages.
• Applications are in areas such as knowledge
  management and electronic commerce.
• Many research projects have been started in the
  EU and US on these topics.
• On-To-Knowledge is one of the first ones breaking
  the ice.