Financial Fraud Prevention-Oriented Information Resources using Ontology Technology

From Wikipedia, the free encyclopedia
Jump to: navigation, search

Financial Fraud Prevention-Oriented Information Resources using Ontology Technology (FF POIROT) explores the use of ontology technology for fraud prevention and detection.[1] The FF POIROT project was an EU fifth framework funded, Information Society Technologies (IST) project (IST- 2001-38248) developed in response to the hundreds of millions of Euros lost every year in the EU due to financial fraud[2] engendered by global digital markets.


The project term was from 1 September 2002 until 31 August 2005, it had a budget of €2,321,388, of which €1,688,821 was funded by the EU.[3] Coordination of the project was led by Dr. Gang Zhao of Vrije Universiteit Brussel's STARLab, Semantic technology and Applications Research Laboratory, they partnered with the following organisations:

Aims of FF POIROT[edit]

The aim of the project was to provide research reports on applying ontology methodology and a computer based, formalized knowledge repository of conceptual information on financial fraud. The project anticipated assisting three distinct user communities, such as the investigative and monitoring bodies to enrich information retrieval for detection. Secondly, the project sought to aid financial professionals in providing an authoritative concept base linked to customized applications, and finally boost law enforcement by facilitating and supporting interoperability between police-orientated query systems.

In order to achieve these aims, the project identified the following requirements/underlying goals:[3]

  • Provide a computationally tractable and shareable knowledge repository for a financial domain available in several languages (Dutch, Italian, French and English).
  • Apply existing technology and develop specific tools to build (potentially standardised) ontologies
  • Commercially exploit the ontology (or parts of it) as a set of Semantic Web services

In meeting the project’s aims, STARLab[5] identified the main innovations of FF POIROT as:

  • Modeling and aligning fairly complex content from interrelated domains
  • Ontology server technology and ontology tools (for alignment, editing, merging)
  • Product definition by incorporating recent scientific insights on ontologies in application specifications
  • Scientific culture (cross-over) by combining computational linguistics with database modeling and web technology

Technical Approach[edit]

The technical approach taken by the project began with a collection of system requirements, which required interviews with users and domain experts. They would then model the financial fraud domain using object-role modeling, and then build an initial ontology base and interpretation layer. The system then mines the corporations followed by human analysis. By using the newly developed adaptable tools they aligned and merged (via collaboration validation) the intermediary results into a final domain ontology. Questionnaires and metrics are then computed by user trials and evaluation the ontology and demonstrators.


Before the FF POIROT project, organisations like CONSOB[4] would detect fraud by conducting a keyword search via publicly accessible search engines (e.g. Google, Yahoo, Microsoft) to flag up websites of interest. The keywords are derived from previous statistical analyses. The websites are subsequently examined manually by Commission’s Officers to identify whether a flagged website is acting fraudulently. These are then ranked in a procedure that replicates a back propagation neural network.[6]

The project ontology incorporates a knowledge system application involving natural language processing[7] and introduces a mechanism through which the footprint of development decision-making is explicitly traceable.[8] The distinguishing feature of knowledge engineering in the project is that it is founded on an approach that uses ontology technology and methodology to manage the modeling and processing complexities in fraud prevention.[9] This use of domain-oriented knowledge resources provides a powerful and flexible solution to identifying fraud.

As financial frauds are crimes defined by the legislative provisions in a given jurisdiction, the texts of laws are the primary sources of ontology modeling in the FF POIROT project. The ontology is used to filter web pages and its natural language text is the medium of forensics combining the domain and language engineering within a text-based ontology.[10]

FF POIROT mines ontology elements from both unstructured and semi-unstructured resources and the Application Knowledge Engineering Methodology (AKEM) used is based on the Developing Ontology-Grounded Methods and Applications (DOGMA) ontology representation framework. The DOGMA framework defines ontology as a set of lexons and their commitments in particular applications.[11]

The lexons are language independent domain specific atomic facts.[12] They represent the types of relationship between two object type and are constructed in the following format: Context, Term1, Role1, Term2, Role2.[11]

Lexons capture the underlying concepts and relationships, while commitments link them to a particular application or task requirement with specific constraints and instantiations.[13] The occurrence of lexons is counted and the average document frequency and average collection frequency are calculated before comparison with predefined classes or keywords. Heuristics are applied to the results and scores of reliability are calculated.

Providing a separate layer that mediates between the ontology base and application instances committing to the ontology makes the process highly efficient and scalable which is vital when handling large databases, a major factor when contending with financial fraud in the digital world. This methodology breaks down the development process into small processes that are then distributed in pipelines of work, which allows for a support structure that accommodates for both independent and collaborative work.

Since the FF POIROT project involves financial fraud in a variety of sub-domains involving a wide range of measures (including international finance, accountancy, tax law, police procedure and evidence handling, general legal and law enforcement knowledge, comparative law knowledge, and knowledge of databases and linguistic descriptions) the ontology is not exclusively confined to the legal domain and incorporates ontologies of finance, evidence and computer science.[14]


Towards the end of the FF POIROT project, user validation reports were gathered assessing its success. CONSOB examined the effectiveness of the ontology-based automatic procedure implemented in the project and identified the following results. When compared with the manual process of keyword searches against pre-determined classes or keywords, the two processes provided different (i.e. no overlapping) results during parallel runs.[15]

It was found in the CONSOB application that both the manual and the ontology-based procedures produced good results in respect of the websites flagged as potentially carrying out fraudulent activities. Finally, the user validation report highlights that the ontology-based application’s processing time was twice as fast as the manual approach’s time. However, this process was automated by way of a bespoke tool in the test for purposes of obtaining wider coverage and avoiding manual intervention.[15]

Recommendations for Improvements[edit]

The user validation report following the CONSOB application of the FF POIROT developed ontology recommended that the ontology-driven application required further improvement in relation to the distinction it made between trusted and suspicious websites.[15]

It was highlighted that the main difficulty for the application was the strong similarities that existed between the content of trusted sites and those that were acting in a fraudulent way, and that a more intelligent utilisation of the knowledge coming from human participants could be used to improve performance. Furthermore, it was recommended in the report that better descriptions of domain key concepts/classes would result in improved performance of the ontology-based application.


  1. ^ Gang Zhao and Robert Meersman Towards a Topical Ontology of Fraud (2006) Lecture Notes in Computer Science 566 Springer Berlin/Heidelberg
  2. ^ Schafer, B The Taming Of The Sleuth–Problems And Potential Of Autonomous Agents In Crime Investigation And Prosecuting 20th BILETA Conference: Over-Commoditised; Over-Centralised; Over-Observed: The New Digital Legal World? April 2005 Queen’s University Belfast pp1
  3. ^ a b Spyns, P FF POIROT: Financial Fraud Prevention-Oriented Information Resources using Ontology Technology (IST-2001-38248) Vrije Universiteit Brussel, Dept. of Computer Science, STAR Lab
  4. ^ a b "". 2013-06-26. Retrieved 2013-09-04. 
  5. ^ "VUB STAR lab". Retrieved 2013-09-04. 
  6. ^ Schafer op. cit. p4
  7. ^ "FF POIROT Ontology Development Portal (IST- 2001-38248) pp7" (PDF). Retrieved October 29, 2009. 
  8. ^ R. Meersman et al. (Eds.): Engineering an Ontology of Financial Securities Fraud OTM Workshops 2004, LNCS 3292, pp. 619. 2004
  9. ^ Supra n6 pp7
  10. ^ Meerman et. al op. cit. pp 610
  11. ^ a b Kerremans, K, Tang, Y, Temmerman, R, Zhao, G: Towards Ontology-based E-mail Fraud Detection (2005) STARLab, Computer Science, Vrije Universiteit Brussel, Belgium at 3.3.1
  12. ^ Richard M Leary, Wim Vandenberghe and John Zeleznikow Joseph Bell Centre for Forensic Statistics and Legal Reasoning School of Law, University of Edinburgh Towards A Financial Fraud Ontology A Legal Modelling Approach Paper submitted to the International Conference on Artificial Intelligence & Law 2003, Edinburgh
  13. ^ Kerremans et al. op. cit. at 3.3.1
  14. ^ Kingston, J, Schafer, B and Vandenberghe, W Towards a Financial Fraud Ontology: A Legal Modeling Approach [2004] Artificial Intelligence and Law 12: 437
  15. ^ a b c "Financial Fraud Prevention–Oriented Information Resources using Ontology Technology IST-2001-38248 User Validation Feedback Report Deliverable D7.1" (PDF). Retrieved October 28, 2009.