Ontotext @ JRC

1
Ontotext _at_ JRC
2
Semantic Web

• The Semantic Web is the abstract representation
  of data on the WWW, based on the RDF and other
  standards
• SW is being developed by the W3C, in
  collaboration with a large number of researchers
  and industrial partners
• http//www.w3.org/2001/sw/
• http//www.SemanticWeb.org

3
Semantic Web (II)

• "The Semantic Web is an extension of the current
  web in which information is given well-defined
  meaning, better enabling computers and people to
  work in cooperation. Berners-Lee et al. 2001
• The spirit
• Automatically processable metadata regarding
• the structure (syntax) and
• the meaning (semantics)
• of the content.
• Presented in a
• standard form
• Dynamic interpretationfor unforeseen purposes

4
Semantic Web Languages

• RDF(S) the next slides
• SHOE, XOL, etc the pioneers
• Topic Maps a metadata language with limited
  impact
• OIL Ontology Interchange Language, the basis of
  the next two http//www.ontoknowledge.org/oil/
• Description Logics-based multilayered language
• DAMLOIL the predecessor of OWL, not to be
  developed
• OWL the W3C standard for Semantic Web ontology
  language, http//www.w3.org/2001/sw/WebOnt/
• Extends RDF(S), but also constraints it
• Has multiple layers (Lite, DL, Full)
• Transitive/symmetric/etc properties,
  disjointness, cardinality restrictions

5
Semantic Web Problems

• Critical mass of metadata is necessary
• Still lack of consensus on many issues (like
  query languages)
• Lack of practices at the proper scale and
  complexity
• Lack of robust Semantic (in our days RDFS)
  repositories
• Should be as flexible, multi-purpose and easy to
  use as HTTP servers and
• As efficient in structured knowledge management
  as RDBMS

6
What are Sirma Ontotext?

• Established in 1992 as a Bulgarian AI Lab.
• Current structure
• Sirma Group International Corp, Montreal, Canada
  
• 8 subsidiary companies the most important ones
  follow below.
• Sirma AI, Sofia
• The RD backbone of the group with two divisions
• Sirma Solutions e-Business, banking, C3,
  e-Publishing, consultancy
• Ontotext Lab Knowledge and Language Engineering.
• EngView Systems, Montreal
• CAD/CAM systems and applications.
• WorkLogic.Com, Ottawa
• Web-based collaboration, workflow, e-Gov.

7
Software Development and Research since 1992

• Track record of success large companies and
  government organizations in US, Canada, Western
  Europe and Bulgaria
• Top-3 Software Company in Bulgaria
• About 70 developers
• ISO 2001 Certificate
• 1999 EIST prize winner

8
Sirma Businesses and Domains

• Diverse business, ranging from COTS products to
  custom projects, consultancy, and outsourcing
  services.
• Major areas
• AI expert systems (beside Ontotext)
• b2b market places
• CAD/CAM (for packaging, quality control)
• e-Government, CSCW, Groupware, Workflow
• Banking
• C3/C4 Systems (military, airport traffic)
• VOIP billing systems
• e-Publishing, Proofing tools.

9
Ontotext Lab

• An RD lab of Sirma for
• Knowledge and Language Engineering
• Research and core technology development for
• knowledge discovery, management, and engineering.
• Specialized for applications in Semantic Web,
  Knowledge Management, and Web Services.
• Aside from the scientific matters, most of us are
  just professional software developers.

10
Leading Semantic Web Technology Provider

• Ontotext is a leading Semantic Web technology
  provider, being
• the developer of the KIM Semantic Annotation
  Platform and
• a co-developer of the GATE language engineering
  platform
• a co-developer of the Sesame semantic repository
  and OWLIM high-performance OWL reasoner
• the developer of the WSMO4J semantic web services
  API
• a partner in the SWAN Semantic Web Annotator
  project.
• Ontotext is part of most of the major European
  research projects in the field the most
  successful Bulgarian participant in FP6.

11
Mission

• A critical mass of research in a number of AI
  areas made efficient KM almost possible.
• the technology on the market is mostly of two
  sorts
• Expensive black boxes
• Academic prototypes
• Our mission is
• To develop and popularize open, skillfully
  engineered tools...
• For Information Extraction and Knowledge
  Management,
• Which considerably reduce the cost for
  implementation and use of KM applications.

12
Major Research Areas

• We focus on building cutting-edge expertise and
  technology in the following areas
• ontology design, management, and alignment
• knowledge representation, reasoning
• information extraction (IE), applications in IR
• semantic web services
• upper-level ontologies and lexical semantics
• NLP POS, gazetteers, co-reference resolution,
  named entity recognition (NER)
• machine learning (HMM, NN, etc.)

13
Academic Technology Partners

• NLP Group, Sheffield University, UK
• Digital Enterprise Research Institute (DERI),
  Institut für Informatik, Innsbruck, Austria,
  andNational University of Ireland, Galway
• Aduna (Aidministrator) b.v., The Nederland's
• Linguistic Modelling Lab.CLPOI, Bulgarian
  Academy of Sciences
• British Telecommunications Plc, (BT), UK.
• Froschungszentrum Informatik (FZI) and Institut
  AIFBKarlsruhe, Germany.

14
Customers

• SemanticEdge GmBH, Berlin, Germany
• QinetiQ Ltd, UK
• Fairway Consultants, UK

15
Research Projects

• We were/are part of a number of FP5 research
  projects
• On-To-Knowledge - the project which invented
  OIL.Ontology Middleware Module and a DAMLOIL
  reasoner.
• VISION - Towards Next Generation Knowledge
  Management.
• OntoWeb - Ontology-based information exchange for
  knowledge management .
• SWWS - Semantic Web enabled Web Services.

16
Research Projects (II)

• FP6 integrated projects that started Jan 2004,
  durations 3 years
• SEKT Semantic Knowledge Technologies. Targeting
  a synergy of Ontology and Metadata Technology,
  Knowledge Discovery and Human Language
  Technology.
• DIP Data, Information, and Process Integration
  with Semantic Web Services.
• PrestoSpace Preservation towards storage and
  access. Standardized Practices for Audiovisual
  Contents in Europe.
• Infrawebs Intelligent Framework for Generating
  Open (Adaptable) Development Platforms for
  Web-Service Enabled Applications Using Semantic
  Web Technologies, Distributed Decision Support
  Units and Multi-Agent-Systems

17
Introduction to Ontologies

• Despite the formal definitions, ontologies are
• Conceptual models or schemata
• Represented in a formalism which allows
• Unambiguous semantic interpretation
• Inference
• Can be considered a combination of
• DB schema
• XML Schema
• OO-diagram (e.g. UML)
• Subject hierarchy/taxonomy (think of Yahoo)
• Business logic rules

18
Introduction to Ontologies (II)

• Imagine a DB storing
• John is a son of Mary.
• It will be able to "answer" just
• Which are the sons of Mary? Which son is John?
• An ontology with a definition of the family
  relationships. It could infer
• John is a child of Mary (more general)
• Mary is a woman
• Mary is the mother of John (inverse)
• Mary is a relative of John (generalized inverse).
  
• The above facts, would remain "invisible" to a
  typical DB, which model of the world is limited
  to data-structures of strings and numbers.

19
Products

• The Ontology Middleware Module (OMM) is an
  enterprise back-end for formal KR and KM
  applications based on Semantic Web standards
• An extension of the Sesame RDF(S) repository
  that adds a Knowledge Control System.
• OMM integration options Built-In, RMI, SOAP,
  HTTP.

20
Products

• BOR a DAMLOIL reasoner.
• Proprietary GATE components
• Hash Gazetteer. A high-performance lookup tool.
• Hidden Markov Model Learner. A stohastic module
  for filtering annotations, disambiguation,
  (etc.,) based on confidence measures.
• The News Collector is a web service, collecting
  and indexing articles from the top-10 global news
  wires
• About 1000 articles/day, annotated and indexed
  using KIM
• Used to validate the heuristics and resources of
  KIM

21
Products (II)

• The KIM Platform (the next slides),
  http//www.ontotext.kim.
• SWWS Studio (http//swws.ontotext.com)
• Semantic Web Service description development
  environment
• Developed in the course of the SWWS project
• Based on WSMO (http//www.wsmo.org)
• WSMO4J (http//wsmo4j.sourceforge.net)
• A WSMO API and a reference implementation
• for building Semantic Web Services applications
• Used in WSMO Studio, (http//www.wsmostudio.org/)
• The basis for ORDI, used in OMWG
  (http//www.omwg.org)
• Used in projects DIP, SEKT, Infrawebs

22
OWLIM

• OWLIM is a high-performance OWL repository
• Storage and Inference Layer (SAIL) for Sesame RDF
  database
• OWLIM performs OWL DLP reasoning
• It is uses the IRRE (Inductive Rule Reasoning
  Engine) for forward-chaining and total
  materialization
• In-memory reasoning and query evaluation
• OWLIM provides a reliable persistence, based on
  RDF N-Triples
• OWLIM can manage millions of statements on
  desktop hardware
• Extremely fast upload and query evaluation even
  for huge ontologies and knowledge bases

23
Scalability Upload and Reasoning
24
Scalability Query Answering

• Q2 Pattern of 12 statement-joins and LIKE
  literal constraint

25
OWLIM under LUMB Benchmark

• The Lehigh Univ. evaluation is one of the most
  comprehensive benchmark experiments published
  recently (ISWC 2004, WSJ 2005)
• Synthetically generated OWL knowledge bases
• The biggest set generated is LUMB(50,0) 6M
  explicit statements
• 14 queries, checking different inferences
• OWLIM on LUMB
• On a desktop machine OWLIM loads LUMB(50,0) in 10
  min
• The only other systems known to load it, does
  this for 12 hours
• All the queries are answered correctly
• Based on this we can claim that
• OWLIM is the fastest OWL repository in the world!

26
JOCI

• Jobs Contacts Intelligence, Innovantage,
  Fairway Consultants
• Gathering recruitment-related information from
  web-sites of UK organizations
• Offering services on top of this data to
  recruitment agencies, job portals, and other.
• JOCI uses KIM for information extraction (IE,
  text-mining)
• JOCI makes use of a domain ontology to
• support the IE process,
• to structure the knowledge base with the obtained
  results, and
• facilitate semantic queries.
• Sirma is shareholder in Fairway Consultants

27
JOCI Dataflow
UK Web Space
Web UI
Information Extraction
KIM Server
Single-Document IE
Semantic Repository
Focused Crawler
Crawler
Classifier
Object Consolidation
Document Store
28
JOCI Vacancy Consolidation/Matching
Consolidated Vacancy
locatedIn
Vacancy 1
Vacancy 2
hasJobTitle
locatedIn
IT Applications Support Analyst
Support Analyst
locatedIn
sub-string
Glasgow
U.K.
Scotland
subRegionOf
subRegionOf
type
type
City
Country
subClassOf
Location
29
JOCI Statistics

• The figures below are indicative and reflect an
  old state of the JOCI system
• The actual figures are to be announced after the
  launch of JOCI
• Web-sites inspected 0.5M
• Web-sites with vacancy announcements 30K
• Extracted vacancies 100K

30
The KIM Platform

• A platform offering
• services and infrastructure for
• (semi-) automatic semantic annotation and
• ontology population
• semantic indexing and retrieval of content
• query and navigation over the formal knowledge
• Based on Information Extraction technology

31
KIM Whats Inside?

• The KIM Platform includes
• Ontologies (PROTON KIMSO KIMLO) and KIM World
  KB
• KIM Server with a set of APIs for remote access
  and integration
• Front-ends Web-UI and plug-in for Internet
  Explorer.

32
The AIM of KIM

• Aim to arm Semantic Web applications
• by providing a metadata generation technology
• in a standard, consistent, and scalable framework

33
What KIM does? Semantic Annotation
34
Simple Usage Highlight, Hyperlink, and
35
Simple Usage Explore and Navigate
36
Simple Usage Enjoy a Hyperbolic Tree View
37
KIM is Based On

• KIM is based on the following open-source
  platforms
• GATE the most popular NLP and IE platform in
  the world, developed at the University of
  Sheffield. Ontotext is its biggest
  co-developer.www.gate.ac.uk and
  www.ontotext.com/gate
• OWLIM OWL repository, compliant with Sesame
  RDF database from Aduna B.V. www.ontotext.com/owl
  im
• Lucene an open-source IR engine by Apache.
  jakarta.apache.org/lucene/

38
How KIM Searches Better

• KIM can match a Query like
• Documents about a telecom company in Europe, John
  Smith, and a date in the first half of 2002.
• With a document containing
• At its meeting on the 10th of May, the board of
  Vodafone appointed John G. Smith as CTO"
• The classical IR could not match
• Vodafone with a "telecom in Europe, because
• Vodafone is a mobile operator, which is a sort of
  a telecom
• Vodafone is in the UK, which is a part of Europe.
• 5th of May with a "date in first half of 2002
• John G. Smith with John Smith.

39
Entity Pattern Search
40
Pattern Search Entity Results
41
Entity Pattern Search KIM Explorer
42
Semantic Metadata in KIM

• Provides a specific metadata schema,
• focusing on named entities (particulars),
• as well as number and time-expressions,
  addresses, etc.,
• everything specific, apart from the general
  concepts.
• Defines specific tasks for generation and usage
  of the metadata which are well-understood and
  measurable.
• Why not metadata about general things
  (universals)?
• It is too complex
• but we leave the door open.
• The particulars seem to provide a good 80/20
  compromise.

43
World Knowledge in KIM

• Rationale
• The ontology is encoded in OWL Lite and RDF.
• provide common knowledge about world entities
• KIM bets on scale and avoids heavy semantics
• minimum modeling of common-sense, almost no
  axioms
• The ontology is encoded in OWL Lite and RDF.
• In addition, a number of rules (generative
  axioms) are defined, e.g.
• ltX,locatedIn,Ygt and ltY,subRegionOf,Zgt gt
  ltX,locatedIn,Zgt
• Axioms of this sort are supported by OWLIM and
  they provide a consistent mechanism for custom
  extensions to the OWL or RDF(S) semantics with
  respect to a particular ontology

44
PROTON

• Name. PROTON is an acronym for
• Proto Ontology
• ex-names BULO (basic upper-level ontology), GO
  (generic ontology)
• not a Russian space rocket ?
• proto used in the sense of primary,
  beginning, giving rise to, vs. first in
  time or oldest
• connotations positive, fundamental, elemental,
  in favour of, even romantic (like a
  science-fiction novel from the 60-ies) ?
• Intended usage. A Basic Upper-Level Ontology like
  PROTON - used for
• ontology population
• knowledge modelling and integration strategy of a
  KM environment
• generation of domain, application, and other
  ontologies.

45
PROTON Design

• Design principles
• domain-independence
• light-weight logical definitions
• Compliance with popular metadata standards
• good coverage of concrete and/or named entities
  (i.e. people, organizations, numbers)
• no specific support for general concepts (such as
  apple, love, walk), however the design
  allows for such extensions

46
Some Figures

• PROTON defines about
• 250 classes and 100 properties
• Providing coverage of most of the upper-level
  concepts necessary for semantic annotation,
  indexing, and retrieval
• A modular architecture, allowing for great
  flexibility of usage and extension
• SYSTEM module - contains a few meta-level
  primitives (6 classes and 7 properties)
  introduces the notion of 'entity', which can have
  aliases
• TOP module - the highest, most general,
  conceptual level, consisting of about 20 classes
• UPPER module - over 200 general classes of
  entities, which often appear in multiple domains.

47
PROTON Ontology Language

• The current version of the ontology is encoded in
  OWL Lite.
• A few custom entilement rules (axioms) are also
  defined for usage in tools that support them, for
  instance
• Premise
• ltxxx, protontroleHolder, yyygt
• ltxxx, protontroleIn, zzzgt
• ltyyy, rdftype, protontAgentgt
• Consequent
• ltyyy, protontinvolvedIn, zzzgt
• Axioms of this sort are interpreted by OWLIM
• PROTON is portable to any OWL(Lite)-compliant
  tool.
• PROTON can be used without such axioms either.

48
Other Standards Relations

• ADL Feature Type Thesaurus and GNS
• the backbone of the Location branch
• on its turn aligned with the geographic feature
  designators, of the GNS database of NIMA
• PROTON is more coarse-grained, taking about 80
  out of 300 types.
• Dublin Core
• the basic element set available as properties of
  protontInformationResource and protontDocument
  classes
• the resource type vocabulary is mapped to
  sub-classes of InformationResource.
• OpenCyc and WordNet consulted and referred to in
  glosses.
• ACE (Automatic Content Extraction) annotation
  types covered.
• FOAF assure easy mapping (e.g. the Account
  class was added).
• DOLCE, EuroWordnet Top, and others consulted to
  various extent.

49
Other Standards Compliance

• Other models are not directly imported (for
  consistency reasons)
• The mapping of the appropriate primitives is
  easy, on the basis of
• a compliant design, and
• formal notes in the PROTON glosses, which
  indicate the appropriate mappings.
• For instance, in PROTON, a protontinLanguage
  property is defined
• as an equivalent of the dclanguage element in
  Dublin Core
• with a domain protontInformationResource
• and a range protontLanguage

50
KIM World KB

• A quasi-exhaustive coverage of the most popular
  entities in the world
• What a person is expected to have heard about
  that is beyond the horizons of his country,
  profession, and hobbies.
• Entities of general importance like the ones
  that appear in the news
• KIM knows
• Locations mountains, cities, roads, etc.
• Organizations, all important sorts of business,
  international, political, government, sport,
  academic
• Specific people, etc.

51
KIM World KB Entity Description

• The NE-s are represented with their Semantic
  Descriptions via
• Aliases (Florida FL)
• Relations with other entities (Person hasPosition
  Position)
• Attributes (latitude longitude of geographic
  entities)
• their proper Class

52
The Scale of KIM World KB
RDF Statements Small KB Full KB
- explicit 444,086 2,248,576
- after inference 1,014,409 5,200,017
Instances
- Entity 40,804 205,287
- Location 12,528 35,590
- Country 261 261
- Province 4,262 4,262
- City 4,400 4,417
- Organization 8,339 146,969
- Company 7,848 146,262
- Person 6,022 6,354
- Alias 64,589 429,035
53
KIM IE Pipeline
54
JAPE Grammars

• Jape grammars are based on the last MUSE version
• Class/instance information included
• Better class granularity in grammars
• Relation recognition grammars - LocatedIn and
  HasPositionWithinOrganization

55
Disambiguation Filtering

• Simple disambiguation (longest match), e.g. San
  Francisco Journal
• Based on the main alias, e.g. Beijing
• By priority of the class, instance or relative
  class priority
• E.g. Brand Microsoft vs. Company Microsoft
  Corp.
• We assign a priority (1-1000) to each class and
  instance
• For pairs of classes we define relative priority
• If the difference between the priorities is
  greater than a certain threshold the possible
  reference to the entity with the lower priority
  is ignored
• Still to be improved

56
KIM Scaling on Data

• The Semantic Repository is based on OWLIM
• In our practical tests we observe perfect
  performance on top of
• 1.2M of entity descriptions
• about 15M explicit statements
• above 30M statements after forward chaining.
• Document and Annotation storage and indexing with
  Lucene
• One million docs, processed on a 1000-worth
  machine
• retrieval in milliseconds.

57
Entity Ranking a sketch for Jan-May 2004
No Instance Label Rank
1 Country_T.5 United States 0.032
4 Country_T.IZ Republic of Iraq 0.011
6 Person_T.51 George W. Bush 0.010
9 Country_T.IS State of Israel 0.006
11 DayOfWeek_T.4 Tuesday 0.005
12 NewsAgency_T.6 The Associated Press 0.005
14 InternationalOrganization_T.13 United Nations 0.005
27 Country_T.CH People's Republic of China 0.004
32 City_T.3068 New York 0.004
36 InternationalOrganization_T.18 European Union 0.004
40 Person_T.115 Ariel Sharon 0.003
43 Country_T.JA Japan 0.003
44 Country_T.UK United Kingdom 0.003
45 CountryCapital_T.93 Baghdad 0.003
58
SWAN/KIM Cluster Architecture

• At present, KIM is used for massive semantic
  annotation in the context of the SWAN and SEKT
  projects
• Here are some of its features
• support for a virtually unlimited number of
  annotators
• centralized ontology storage and querying
• centralized meta-data (annotations) and document
  storage, indexing, and querying
• support for multiple crawlers (or other data
  sources)
• dynamic reconfiguration of the cluster (e.g.
  staring new crawlers or annotators on demand).

59
SWAN/KIM Cluster Console
60
SWAN Project Semantic Web Annotator

• Large Scale Annotation of human language for the
  Semantic Web using Human Language Technology
  (HLT).
• Hosted by DERI (NUIG, Galway) and involves also
• GATE team (from the Sheffield University's NLP
  Group) and
• Ontotext Lab.
• For more details take a look at
  http//deri.ie/projects/swan/
• The current status
• KIM Cluster of 7 servers in DERI
• Above 0.5TB shared storage
• 6 AMD64 Opterons, 6 Xeons, 36GB RAM

61
CoreDB Name and Goals

• CoreDB is a component of KIM
• Stands for Co-Occurrence and Ranking of Entities
  DB
• In a nutshell, it is designed to allow fast
  queries of the sort
• Q1 the number of appearances of UK in
  documents during Jan 2005
• Q2 all people co-occurring with John Smith and
  some bank institution in documents from the
  second half of 2003
• Q3 Q2 where the documents contain fraud and
  the name of the institution contains capital

62
CoreDB Functionality

• It allows asking in a structured manner for
• The number of references to entities in a
  (sub-)set of documents
• The entities, which co-occur together with other
  entities
• Entities can be constrained by
• Class (and its sub-classes)
• Keyword/token in one of its names/aliases/labels
• Documents can be constrained according to DC-like
  features
• Date (range could be any date in the doc)
• Type (exact match could be any string)
• Authors
• Title and Sub-title
• Keyword/token in the content, authors or the
  title fields

63
The Scale of Ambition

• The major point is to allow such queries in
  efficient manner over data with the following
  cardinality
• 106 entities/terms
• 107 documents
• 102 entities occurring in an average document
• This means managing and querying efficiently 109
  entity occurrences
• We had tested the current implementations with
  107 occurrences and it answers the basic
  queries in milliseconds.

64
CoreDB Applications

• Detection of associative links between
  entities, based on co-occurrence in documents
• It is an alternative of the detection of strong
  links based on local context parsing
• Ranking, measuring popularity, of an entity over
  a set of documents
• The ranking is as good/relevant/representative as
  the set of documents is
• Computing timelines (changes over time) for
  entity ranking or co-occurrence
• How did our popularity in the IT press changed
  during June (i.e. What is the effect of this
  1.5MEuro media campaign ?!?)
• How does the strength of association between
  organization X and RDF changes over Q1 ?

65
Implementation

• It is a new component in the architecture of KIM
• Having an API (part of the KIM API), allows
  different implementations
• There are now a couple of RDBMS-based
  implementations
• Derby (free, open-source, 100 Java, was
  Cloudscape from IBM)
• ORACLE (v. 10g)
• The Derby implementation does not allow for
  efficient searches involving keywords
• The ORACLE implementation is used also for
  FTS-style indexing of the document contents
• Makes possible efficient combination of semantic
  and keyword search (which is already available
  through the SemanticQuery API)
• In both RDBMS implementations
• Part of the ontology and the KB are replicated
• Same with part of the document and index related
  information

66
Ontotext Facts

• Founded year 2000
• 14 employees (permanent, without the shared
  personnel and associates)
• Daily statistics for http//www.ontotext.com,
  over 150 visits 2000 hits
• Number of scientific publications above 30
• Number of projects running 9
• More than 20 partners we directly cooperate with
  on projects
• Average age about 28
• Number of servers per developer 0.7

67
Ontotext Lab

• Robust Technology
• and Professional Services for
• Knowledge and Language Engineering
• http//www.ontotext.com