Terminology and Knowledge Engineering in Fraud Detection

1
Terminology and Knowledge Engineering in Fraud
Detection
Koen KerremansRita Temmerman
Gang Zhao
Centrum voor Vaktaal en Communicatie
(CVC)Department of Applied LinguisticsErasmusho
geschool Brusselhttp//cvc.ehb.be
Semantics Technology and Applications Research
Laboratory (STAR Lab)Department of Computer
ScienceVrije Universiteit Brusselhttp//www.sta
rlab.vub.ac.be
2

• How are terminology and knowledge engineering
  used in the fight against financial fraud?
• How to organise terminology and knowledge
  engineering methods into a development process of
  technological solutions in the fight against
  financial fraud?

3
General outline

• FF POIROT
• Aims
• Cases
• Partners
• Methodologies
• AKEM (knowledge engineering)
• Termontography (terminology engineering)
• Interaction of methodologies
• Future work
• Conclusion

4
FF POIROT
Financial Fraud Prevention Oriented Information
Resources using Ontology Technology

• Aims
• Apply Semantic Web technology to fraud detection
  and prevention, thereby showing the potential of
  ontologies in these areas
• Construct multililingual terminological as well
  as formal knowledge repositories covering the
  domains of interest
• Propose methods and guidelines in terminology and
  knowledge engineering
• Develop new and/or improve existing tools to
  support terminology and knowledge engineering

5
FF POIROT cases

• VAT carousel fraud
• VAT fraud in which fraudsters sell goods at VAT
  inclusive prices and disappear without paying the
  VAT paid by their customers to the tax
  authorities
• Companies unwittingly involved in this type of
  fraud can be held responsible for the missing
  VAT
• Each company has to find out whether or not it is
  safe to do business with a trader from another
  EU country
• On-line investment fraud
• the selling of overpriced or worthless shares,
  bonds, or other financial instruments to the
  general public
• In Italy, Consob searches suspicious websites via
  traditional search engines such as Google,
  Altavista,

6
FF POIROT

• Use of the ontology
• Knowledge management of fraud investigative
  expertise
• Information exchange between investigative bodies
• Automation of parts of monitoring or
  investigative procedures with knowledge-based
  applications (e.g. information extraction)
• Use of multilingual terminology
• Dicionary purposes
• Multilingual keywords in information extraction
• Explanation of reasoning in natural language
• Knowledge resource consulted during ontology
  development

7
FF POIROT partners
8
AKEM

• Application Knowledge Engineering Methodology
• Development cycle
• Knowledge scoping (result stories)
• Knowledge analysis
• Ontology development
• Deployment

9
AKEM

• Based on DOGMA
• Developing Ontology-Guided Mediation for Agents
• Ontology a set of lexons and their commitments
  in particular applications
• Lexon a grouping element stored in a lexon base
  and composed of terms and roles
• ltContext, Term_1, Role_1, Term_2, Role_2gt

10
AKEM

• Why Application Knowledge Engineering
  Methodology?
• There is a need to organise a geographically
  distributed, multidisciplinary team of domain
  experts, knowledge analysts and engineers in a
  methodical traceable development cycle
• There is a need to examine how knowledge can be
  extracted from different perspectives on fraud to
  improve the quality of the fraud ontology

11
AKEM
H1

• An example of a legal view Wigmore chart
• Blue hypothesis
• Red claim
• Purple evidence
• Green fact

1
1.1
1.2
1.3
1.1.1
E1.2.1
E1.3.1
E1.1.1
F1.1.1.1
F1.1.1.2
12
AKEM

• H1 Public offer of company X is unlawful
• 1.1 X solicits investors on the WWW
• E1.1.1 X manages website that solicits investors
• F1.1.1.1 Website states name X
• F1.1.1.2 Registration details indicate X as
  registrant of website
• 1.2 No advance notice of solicitation to Consob
• E1.2.1 X did not give a notification to Consob
  regarding public offer to purchase
• 1.3 No related prospectus filed with Consob
• E1.3.1 X did not draft or file a prospectus with
  Consob regarding public offer to purchase

13
AKEM

• Extraction of knowledge constituents and
  abstraction into production rules, allow
  knowledge modellers to identify and organise the
  abstract concepts and relations into a lexon base
• Example

14
Termontography

• a terminological approach in which
  (multilingual) terminological knowledge,
  retrieved from texts, is structured according to
  a framework of knowledge (i.e. categorisation
  framework)
• Why Termontography?
• Terminographers need a common reference framework
  to scope the terminology work
• There are significant commonalities between
  terminology compilation and text-based ontology
  development
• In our view a terminological analysis can
  contribute to the formalisation of a given domain

15
Termontography
Search phase (3)
(mono- or multilingual) domain-specific corpus
first version of termontological database
Ontology
Dictionary
Refinement phase (4)
(mono- or multilingual) termontological database
Information gathering phase (2)
TSR categorisation framework
Verification phase (5)
Domain- experts
Validation phase (6)
Knowledge Analysis phase (1)
16
Interaction of methodologies
AKEM
TERMONTOGRAPHY
KNOWLEDGE SCOPING
17
Interaction of methodologies

• Knowledge scoping
• Developing terminological resources and
  ontological repositories requires above all an
  insight in the domain of interest
• Domain experts can support the knowledge
  acquisition process by pointing out the relevant
  categories/topics (given the envisaged
  tasks/applications)
• Example Transactions for which no VAT is
  required

18
Interaction of methodologies

• Terminology base ? ontology development
• Rationale
• The AKEM extraction task seeks for basic semantic
  elements and follows linguistic units in natural
  language texts
• Experience shows that ontology engineers resort
  from time to time to terminological resources for
  background information or exact definitions
• Characteristics of terminological analysis
• Special emphasis on documenting semantic contexts
  by means of textual contexts
• Entries also include linguistic semantic
  descriptions such as agent-predicate-patient/recip
  ient links and cross references among items of
  contents

19
Interaction of methodologies
20
Interaction of methodologies
21
Interaction of methodologies

• Consequences
• Productivity of ontology engineers is improved by
  suggestions from terminologists, who examine the
  same knowledge resources
• Terminography adds a linguistic viewpoint to
  application-specific modeling
• During ontology development, multilingual
  terminological information can help discover
  semantic gaps across languages due to social
  and cultural differences and facilitate consensus
  building in a multilingual team of developers

22
Future work

• How to represent meaning variations of
  lexicalisations referring to the same category?
• E.g. An event on which VAT is to be paid
• Irish legislation chargeable event
• VAT will be due on the date the invoice is issued
• UK legislation chargeable event
• VAT will be due no later than the 15th day
  following the month in which the supply takes
  place
• French legislation fait générateur
• VAT will be due at the moment the goods are
  supplied

23
Future work

• Termontography workbench

24
Conclusion

• We have shown how terminology and knowledge
  engineering is used in the fight against
  financial fraud
• We have shown how the methods of experts with
  different backgrounds have been translated into a
  coherent and traceable workflow

25
Conclusion
26
Conclusion

• Project
• http//www.ffpoirot.org
• Partners
• STAR Lab http//www.starlab.vub.ac.be
• CVC http//cvc.ehb.be
• JBC http//www.cfslr.ed.ac.uk/
• RACAI http//www.racai.ro
• KS http//www.knowledgestones.com/
• LC http//www.landcglobal.com/index.php
• Consob http//www.consob.it/main/index.html
• VAT_at_ http//www.vatat.com