=Paper=
{{Paper
|id=Vol-1898/paper9
|storemode=property
|title=Ontology Based PEP Identification
|pdfUrl=https://ceur-ws.org/Vol-1898/paper9.pdf
|volume=Vol-1898
|authors=Andrejs Gaidukovs,Pavels Grjozs,Girts Kebers
|dblpUrl=https://dblp.org/rec/conf/bir/GaidukovsGK17
}}
==Ontology Based PEP Identification==
https://ceur-ws.org/Vol-1898/paper9.pdf
Ontology Based PEP Identification
Andrejs Gaidukovs1, Pavels Grjozs 1, Girts Kebers2
1 Institute of Applied Computer Systems, Riga Technical University, Latvia
{andrejs.gaidukovs; pavels.grjozs}@edu.rtu.lv;
2 SIA “Lursoft IT”, Latvia
{kebers}@lursoft.lv
Abstract. In many countries financial organizations have obligations to identify
politically exposed persons. The aim of the research work presented in this paper
is to develop the solution that gives an opportunity to identify politically exposed
persons using information from the web and several state registers. In particular,
the paper focuses on one of the possible solutions that is based on the ontology
with the dedicated inference rules. This paper proposes the ontology for
identification of politically exposed persons and discusses the problems faced
during the development an application of the ontology, showing the limitations
of the ontology based approach and pointing to some potential solutions of the
identified problems.
Keywords: ontology, politically exposed person, PEP, identification,
compliance, anti-money laundering, countering the financing of terrorism
1 Introduction
Politically exposed person (PEP) is defined as an individual who is or has been
entrusted with a prominent public function. Due to their position and influence, it is
recognized that many PEPs are in positions that can be abused for the purpose of
committing money laundering offences and related predicate offences, including
corruption, as well as conducting activity related to terrorist financing [1].
This is a research-in-progress paper to report on the preliminary results of the
ongoing research regarding the use of the ontology based approach in PEP
identification. The project is held in Latvia, thus, there are some country specific
regulations incorporated in the solution that can be different in other countries.
The following are three main sources that regulate and describe the rules for PEP
identification:
─ Latvian Anti-Money Laundering / Countering the Financing of Terrorism
(AML/CFT) law [1];
─ recommendations [2] issued by Financial and Capital Market Commission
(FCMC) that regulates financial organizations’ operations and compliance in
Latvia, among other, on how to identify politically exposed people;
2 Andrejs Gaidukovs, Pavels Grjozs, Girts Kebers
─ Financial Action Task Force (FATF) [3] – recommendations of international
governmental organization that stipulates 40+9 policies on AML/CFT .
Politically exposed persons are divided into three main categories:
1. A PEP itself – individual who is or has been entrusted with a prominent public
function [3];
2. Family members of a PEP;
3. Persons who are closely related to a PEP – the person that has co-owned private
companies with the PEP or has business relations with the PEP.
There are several differences between definitions and rules if various sources of
regulations are compared. For instance, time period when the person is considered as
politically exposed even after he/she has left PEP indicative position from FCMC
definition is 12 months. If the person was PEP then person stays PEP forever by FATF
definition. For the purposes of this paper, the FCMC definition of PEP status expiry is
used i.e.– 12 months. Another example of unclear definition is business relation
definition. Not all deals have to be taken into consideration, as mentioned in all
definitions, but there is no clear threshold of the relevant deal size or any other
parameters stated in the above-mentioned regulations.
The following research method was used in the study: 1) possible data sources
relevant for PEP status identification were analyzed and types of data were defined and
structured; 2) related work on PEP identification was studied; 3) preliminary PEP status
identification ontologies were built and tested; 4) after the analysis of the test results
the final ontology was built; 5) the graph database was used to solve PEP identification
challenges that were not met by the ontology.
The paper is organized as follows. In Section 2, the current situation in PEP
identification solutions is described. Section 3 proposes and discusses ontology based
solution for PEP identification. Section 4 concludes the paper.
2 State-of-Art in PEP Identification
Identification of PEPs is a significant issue in the field of finance and national
security. CaseWare Analytics [4] describes a solution for the global identification of
PEPs. CaseWare Analytics study offers the division of PEP status into four classes
depending on the probability (risk), as well as provides an insight into the development
of a PEP identification system. Possible solutions for the issue of PEP identification
and associated data retrieval risks are provided in several patents: Mark. A. Schiffer’s
patent "Method of Ranking Politically Exposed Persons and Other Heightened Risk
Persons and Entities" [5] , David Lawrence’s patent “Automated Political Risk
Management” [6] , patent “Evaluating Customer Risk” [7]. All these methods use
different information acquisition methods from several data sources. To join these data
sources, bodies of related notions are used, for example, controlled dictionaries,
concept maps, and ontologies.
Ontologies are used in Anti-Money Laundering systems. One of examples for the
use of ontologies is an article by Dr. Jerry A. Smith [8] showing a money laundering
Ontology Based PEP Identification 3
scheme from a perspective of an ontology. The scheme shows concepts understandable
to both a person and interconnected system that have to be able to operate in this domain
of knowledge.
There are number of software products available on the market for financial
institutions and insurance companies for maintaining compliance with anti-money
laundering requirements, for example: Actico Name Matching Customer [9], FICO
TONBELLER IT-based Anti-Money Laundering [10], IBM AML compliance solution
[11]. All these solutions use external monitoring lists for comparing with the customer
lists. This process includes comparing of similar names, aliases, and spelling variations
of names, as well as birthdates, nationality and domicile of the persons concerned
against monitoring lists. Official lists such as the EU list or OFAC list can be used for
monitoring, as well as those offered by commercial services. The purpose of the
solution presented in this paper is to provide a commercial service that contains actual
list of PEPs.
3 Proposed solution
Proposed solution analyzes data that is available on the web and in state registers.
First, the available data sources were analyzed and Resource register was developed for
these data sources. Then the data sources in Resource register were rated by data
reliability and credibility. Further, Resource register was used for matching the
attributes of the data objects available in various data sources. Actual data was retrieved
from data sources using crawlers. The ontology was used for person PEP status
identification. The proposed ontology (Fig. 1) is created using Protégé [12] tool. The
proposed ontology has three major classes “Person”, “Position” and “Organization”.
Fig. 1. Proposed PEP identification ontology in Protege
Class “Person” has two subclasses: “PEP” and “ClosePerson”. Two subclasses are
needed to have a possibility to stop the reasoning. When PEP is identified, the list of
4 Andrejs Gaidukovs, Pavels Grjozs, Girts Kebers
close persons is defined. (Fig. 2) If only one class “PEP” was used then the status PEP
would be assigned to each person from ClosePeople list in step 3 (Step 3: For
i=1 to M do ClosePeople[i]=PEP). Then assigned people would be searched
for persons close to PEP and identified persons would again be granted PEP status.
Such algorithm would infer that all people are PEPs. If class “ClosePerson” is used then
ClosePerson status is assigned to each person from the list in step 3 (Fig. 2) and
inference stops there. Another option would be using three subclasses as PEP
categories, but that would require introducing additional class “FamilyMember”. This
class would then have the same properties and behaviors as class “ClosePerson”.
Identification of both subclass instances is the same (and, thus, the information
processing for financial institution is also the same). Therefore, there is no need to have
three subclasses and it is sufficient to have two subclasses to identify the PEPs.
Step 1: PEPn identified
Step 2: Identify the list and search for PEP n closed people
ClosePeople[1..M]
Step 3: For i=1 to M do ClosePeople[i]=ClosePerson
Fig. 2. PEP identification steps
Class “Position” is divided into two subclasses: PEPindication and Level.
“PEPindication” describes possible positions: PEP and nonPEP positions. “Level”
describes management level and consists of “Middle management” and “Top
management”. The research focuses on top management positions because PEPs are
from top management level. Top management level has three subclasses:
1. OnlyInPrivateOrg – the class that contains positions that refer only to private
organizations. For instance, position “owner” is only possible in private organization
as the State organisations would not have any private individuals as owners (i.e. state
organisations are owned by state).
2. InStateOrg – the class that contains positions referring to public organizations. For
instance, minister, deputy, judge – these positions can be held by persons that work
only in public sector.
3. PossibleInStateOrInPrivateOrg – a person with such position can work either in
private sector or in public sector. For instance, board member or director. State
owned organizations have management boards and board members, as well as state
agencies hire directors.
The class “Organization” has to identify the type of organization and four subclasses
are possible: Political, StateOwned, Public or Private. Table 1 shows possible
combinations when person is politically exposed based on the position held and
organization type he/she works for. (“Yes” in the table means that the person is
politically exposed.)
Ontology Based PEP Identification 5
Table 1. PEP positions matrix
Position available Position available in
Position available only
only in private private and in public
in public sector
sector sectors
Private organization Not available No No
Public organization Yes Not available Yes
PEP identification ontology’s object properties are described in Table 2. Object
property “hasRelation” has tree sub-properties: “isBusinessCoowner”,
“isBusinessPartners”, “isFamilyMember”. Sub-properties have the same domain and
range as parent property.
Table 2. PEP identification ontology’s object properties
Object properties Domain Range
hasJob Person Position
hasRelation Person Person
worksFor Person Organization
The following rules were incorporated into the Ontology:
1. Class “InStateOrg” is equivalent to class “PEPIndicativePositions”:
InStateOrgPEPIndicativePositions
2. Person who has job from PEP Indicative positions list is politically exposed:
x (hasJob(x, PEPIndicativePositions)) → PEP(x)
3. Person who has position available in private or in public sectors and works for
state-owned company is politically exposed:
x (hasJob(x, PossibleInStateOrInPrivateOrg)
worksFor(x, StateOwned)) → PEP(x)
4. Every person who is a family member or has a co-owned business or has a
business relationships with the PEP is close to PEP person:
x,y ((Familymember(y,x) ∨ BusinessCoowner(y,x) ∨
BusinessPartner(y,x)) PEP(x)) → ClosePerson(y)
Created rules and classes were tested successfully. Ontology is consistent and
reasoning works correctly.
In further development the challenge to incorporate the period when a person is
considered as a PEP after the person has left the PEP indicative position was addressed
(in this study this period is 12 months). It was needed to add to object property time
period as attribute and later to define rules that use this attribute in inference process.
The only discovered possibility to implement this was to use additional class “Interval”
and use instances of this class in reasoning. Such solution would mean that for every
person–position relation class “Interval” instance must be defined. The solution would
have as many Interval instances as the number of aforementioned relations.
6 Andrejs Gaidukovs, Pavels Grjozs, Girts Kebers
As described earlier, another challenge is that PEP identification regulations have no
clear guidance on possible deal parameters and their threshold value. During the project
there were defined some attributes based on deal value and available related
information in Latian law. While in experiments different limits were tested, the
following two potential limits were investigated for the practical usage:
1. Limits defined in Latvian AML/CFT law [1] – Euro 15 000. If a financial
organization notices money transaction with sum equal or larger than Euro 15000,
the financial institution has an obligation to pay higher attention to and control
customers who have made this money transaction;
2. Limit defined in Latvian public procurement law [13] – Eur 10000. This limit is
starting sum when the state organization has to issue public procurement procedure.
In order for the proposed ontology to evaluate the deal criteria for the identification
of persons closely related to a PEP based on their business transactions, the proposed
ontology needs to be extended to incorporate additional object property
“isBusinessPartners” and attribute “DealValue”. Possibility to add the attribute to the
object property and using of this attribute in the reasoning process was not found.
Another problem faced during development of ontology was of ontology was as
follows. Fig. 3 shows a screenshot from Protégé tool [12] with person’s object
properties. If the person has one position and one employer the inference works
correctly. However, if the person has two positions in two types of organizations, the
inference result is not satisfactory. In the example in Fig. 3, the person works in L-
energo (state owned company) as a project manager and holds board member position
in ABC Ltd (private company). In real life the person is not a PEP, but Protégé reasons
that this person is the PEP because the person works in the state-owned company on a
top management level. Actually all instance properties are defined as independent rules.
But in this study there is the need to combine two rules into one.
Fig. 3. Person with different positions
After investigation, all three above-mentioned problems were identified as N-ary
relation problems. Possible solutions of this problem are described in [14]. All solutions
offer to add a class and define the range as instances. However, in this study there is no
possibility to define the range of time intervals or moments for every query that is
processed. Possible solution for property chains is described in [15], but we did not
succeed to implement it in Protégé. Other ontology editing tools were investigated in
order to satisfy the needs of current ontology. Only solution that handles n-ary relations
is x-Protégé [16]. This is add-on to Protégé tool. But after many experiments and due
to lack of x-Protégé tool manual there still was no result.
Ontology Based PEP Identification 7
Experiments were switched from ontology to graph database. The graph database
tool OrientDB [17] was used. OrientDB tool allows definition of attributes for
relationships and there are no restrictions on the number of attributes for relationships.
The only limitation is that only one attribute can be visualized for a relationship.
Defined attributes can be used in querying. Fig. 4 shows realization of “hasDeal”
relation were deal value of 20 000 Euro is defined. The time attribute is added to the
relation “hasJob” that depicts the time moment when a person starts working as CEO.
Also the chain of relations hasJob->worksFor can be defined in OrientDB. Thus, the
combination of ontology approach with graph database allows solving all three above
mentioned challenges.
Fig. 4. Relations with attributes in OrientDB tool
4 Conclusion
This paper has been developed with an aim to evaluate the possibility of using
ontology approach for the identification of politically exposed persons (PEPs), their
family members and close associates.
To ensure the compliance of the offered solution with requirements set forth in laws
and regulations of the Republic of Latvia; laws and regulations defining the PEP status
have been analyzed within the framework of this paper, thus defining the amount of
search tasks and necessary data.
The studies in the field of identification of PEP status have been analyzed in this
paper as well. The topic of PEPs is shortly covered in scientific articles; however,
several commercial products were analyzed and patent information was studied.
The ontology for PEP identification has been developed and tested. Based on the
position held by a person, the ontology applies the laws of logical reasoning to conclude
whether the person is to be assigned with a PEP status or not. Likewise, the ontology
infers whether a person is closely related to a PEP in cases when this person has close
relations with the relevant PEP.
During ontology development there were number of limitations identified for this
method. It was not possible to add an attribute to the relation. Possible solutions
described in literature were not suitable for PEP status identification. After many
experiments with the ontology, as an alternative, the graph database method was tried
and gave satisfactory results. Graph database OrientDB tool supports attribute
attachment to a relation. Also this tool supports relation chains. The graph database
method is suitable to be combined with the developed ontology.
8 Andrejs Gaidukovs, Pavels Grjozs, Girts Kebers
The ontology shows good results when implementing reasoning. When defining
equivalent classes and object properties, the inference works correctly using these
defined languages. The research work on combining the ontology with other methods
such as information fusion and agent technologies is continuing for finding the most
effective solution for PEP identification not only at the national, but also at the
international levels.
Acknowledgement. The research has been supported in part by the funding from the
research project "Competence Centre of Information and Communication
Technologies" of EU Structural funds, contract No. 1.2.1.1/16/A/007 signed between
IT Competence Centre and Central Finance and Contracting Agency, project No. 1.14.
References
1. Latvian Anti-Money Laundering / Countering the Financing of Terrorism (AML/CFT) law,
https://likumi.lv/doc.php?id=178987.
2. FKTK, http://www.fktk.lv/.
3. FATF: FATF Guidance. (2013).
4. Webinar: Reduce Time Spent Screening Politically Exposed Persons | CaseWare Analytics,
https://www.casewareanalytics.com/events/webinar-reduce-time-spent-screening-
politically-exposed-persons.
5. Schiffer, M.: Method of ranking politically exposed persons and other heightened risk
persons and entities. (2008).
6. Lawrence, D.: Automated political risk management. (2002).
7. Evaluating customer risk. (2006).
8. Deep Learning Intelligence Platform – Addressing the KYC AML Counter Terrorism
Financing Challenge – Data Scientist Insights,
https://datascientistinsights.com/2014/12/17/deep-learning-intelligence-platform-
addressing-the-kyc-aml-terrorism-financing-challenge/.
9. Smarter Decisions with ACTICO, https://www.actico.com/en.
10. FICO TONBELLER - Anti-Money Laundering | Fraud Prevention | FATCA Compliance |
Know Your Customer, http://www.tonbeller.com/en/.
11. IBM AML compliance solution, https://www-01.ibm.com/common/ssi/cgi-
bin/ssialias?htmlfid=IIS03009USEN.
12. Protégé, http://protege.stanford.edu/.
13. Public procurement law, https://likumi.lv/doc.php?id=133536.
14. Defining N-ary Relations on the Semantic Web, https://www.w3.org/TR/swbp-n-
aryRelations/.
15. Hitzler, P., Krötzsch, M., Rudolph, S.: Foundations of Semantic Web technologies. CRC
Press (2010).
16. Willms, C., Krieger, H.-U., Kiefer, B.: ×-Protégé An Ontology Editor for Defining Cartesian
Types to Represent n-ary Relations.
17. OrientDB - Distributed Multi-Model and Graph Database, http://orientdb.com/orientdb/.