Entity Identification Problem in Big and Open Data
J. G. Enríquez
1
, Vivian Lee
2
, Masatomo Goto
2
, F. J. Domínguez-Mayo
1
and M. J. Escalona
1
1
Department of Computer and Language Systems, University of Seville, Seville, Spain
2
Fujitsu Laboratories of Europe, Hayes, Middlesex, U.K.
Keywords: Software Engineering, Big Data, Open Data, Entity Identification, Intelligent Reconciliation, Virtual Graphs.
Abstract: Big and Open Data provide great opportunities to businesses to enhance their competitive advantages if
utilized properly. However, during past few years’ research in Big and Open Data process, we have
encountered big challenge in entity identification reconciliation, when trying to establish accurate
relationships between entities from different data sources. In this paper, we present our innovative Intelligent
Reconciliation Platform and Virtual Graphs solution that addresses this issue. With this solution, we are able
to efficiently extract Big and Open Data from heterogeneous source, and integrate them into a common
analysable format. Further enhanced with the Virtual Graphs technology, entity identification reconciliation
is processed dynamically to produce more accurate result at system runtime. Moreover, we believe that our
technology can be applied to a wide diversity of entity identification problems in several domains, e.g., e-
Health, cultural heritage, and company identities in financial world.
1 INTRODUCTION
In the current Big Data era, quintillions bytes of data
are produced everyday. Better utilization of Big Data
has shown great benefits to organizations for accurate
and faster decision-making, thus enhancing business
performance and competitive advantage. However,
the 3Vs’ nature of the Big Data, e.g. velocity, variety,
and volume, have also presented grand challenges to
most of the traditional information systems, in terms
of data processing, integration, and analysis
(Manyika et al, 2011).
In addition to the Big Data, Open Data has also
emerged as another hot topic recently. The idea is that
certain data should be freely available to everyone,
without any restrictions. Open data has gained more
popularity in the recent years, especially with the new
initiatives of open-data government such as Data.gov
and Data.gov.uk (Official websites of government:
data.gov and data.gov.uk, 2014). The power of Big
and Open Data is enormous, if one can properly
unlock and discover the insight. However, one of the
major hurdles during the chained process - from
receiving the Big and Open Data to applying proper
data analytic tool is that, data are always siloed.
Further more, even data are integrated from different
sources, reconcile the information to refer to the same
entity, proved to be a big challenge, the reasons being
different data sources issue different local identity to
the entity that may already has other identity in other
systems, in the mean time, different business domain
has proprietary way of defining local identity without
a standard mechanism.
In this paper, we present our work to tackles this
issue. We introduce our technical background in
section II, and give out scenarios where entity
identification is the key issues in section III. Section
IV further explains our technology in details, and we
draw conclusions to our work in section V. Finally,
we present our future road maps in section VI.
2 BACKGROUND
2.1 Big Data Platform
Fujitsu Laboratories of Europe Ltd. has been
conducting Big Data and Open Data related research
since the emergence of the phenomena. Over the
years, we have matured our Big Data platform -
BigGraph for data integration, storage, and
processing. Empowered by Linked Data (Berners-
Lee, 2006) technology, the BigGraph platform is able
to efficiently extract, and integrate Big Data from
heterogeneous sources in variety of types, into a
common analysable format. Figure 1 is a simple
illustration of the platform:
404
G. Enríquez J., Lee V., Goto M., J. Domínguez-Mayo F. and J. Escalona M..
Entity Identification Problem in Big and Open Data.
DOI: 10.5220/0005470704040408
In Proceedings of the 17th International Conference on Enterprise Information Systems (ICEIS-2015), pages 404-408
ISBN: 978-989-758-096-3
Copyright
c
2015 SCITEPRESS (Science and Technology Publications, Lda.)
Figure 1: Fujitsu Big Data Platform.
At the time of writing, the platform is able to handle
data sources that contain structured, semi-structured,
and un-structured data types.
It is also worth explaining that Linked Data relies
on two technologies that are fundamental to the Web:
Uniform Resource Identifiers (URIs) (Berners-Lee,
2005) and the HyperText Transfer Protocol (HTTP)
(Fielding, 1999). While Uniform Resource Locators
(URLs) have become familiar as addresses for
documents and other entities that can be located on
the Web, Uniform Resource Identifiers provide a
more generic means to identify any entity that exists
in the world.
2.2 Virtual Graphs
Graph technology is a nature solution to handle Big
Data, especially for modelling relationships between
entities. The variety of graph algorithms, for example,
Dijkstra, A*, and Kruskal offers great flexibility in
different situations. Theoretically, graphs can be
represented in two ways: explicitly and implicitly. An
explicit graph is a collection of elements that can be
completely stored in memory, which means each
vertex and edge of the graph is fixed at the time it is
stored. On the other hand, an implicit graph is a graph
that cannot be stored in memory because of its large
size (Mondal and Deshpande, 2012).
At the University of Seville, we have been
conducting research into implicit graph for many
years; we have formally named our concept Virtual
Graphs. The ability of building the graph at runtime,
allows us to build different solutions to tackle many
business scenarios, where the fixed predefined data
model cannot cope with the extensibility or the
unpredictable availability of the data sources.
3 INSPIRATION
In a recently published white paper, James and Nigel
stated “managing identifiers is easier in a closed
system. The web has many advantages, but it presents
challenges for identifiers because of its vast scale and
their ad hoc usage” (Powell and Shadbolt, 2014). In
real life, we clearly see problems in different domains
that require entity identifications to be reconciled.
Some detailed examples are explained as follows:
Management of cultural heritage information is a
big issue in the Andalusian region (Spain), because
there are lots of monuments that need to be uniquely
identified. In addition, there are different kinds of
relationships between monuments. For instance, a
monument can be associated with another one, or it
can be grouped within a set of monuments.
Furthermore, a monument can contain lots of
artworks, and these artworks can be associated to
other existing monuments. Assigning/reconciling
identifications and build relationships between
monuments requires a very comprehensive system
that current does not exist yet.
In e-Health, to be able to accurately identify
patients is a big challenge, since it requires the
advanced solutions to allow different clinics to
exchange healthcare information in a reliable and
EntityIdentificationProbleminBigandOpenData
405
secure way. Moreover, for those organizations that
exchange healthcare information without using a
common unique identifier or reconciled identity
information, the successfulness of the information
linkage is relying entirely on the accuracy and
completeness of the key demographic data.
Another typical example of the identification
reconciliation issue is the identifiers of the companies
worldwide. In financial world, there are many
different regulators who produce variety of financial
reports for companies all over the world. Each of
these regulators issues unique identifiers within their
own system to the companies. However, currently
there is no any good system that can help users to
integrate all the identifiers issued by different
regulators, and relate them to the same company
entity. Thus making integration of reports from
different regulator almost impossible.
4 OUR IDEA
4.1 EIDER Model
Motivated by the problems and challenges described
in the previous section, we feel the urge to develop a
platform that is able to solve the entity identification
problems, which allows organizations to gain even
more benefits when utilizing Big and Open Data
hence the birth of EIDER platform.
Based on the mature BigGraph platform
previously mentioned, EIDER platform is further
enhanced to tackle the entity identification problem.
The system architecture of EIDER is illustrated in the
figure 2.
The system consists of three main functionalities:
Entity Extraction
Entity Reconciliation
Data Integration
Each of the functionalities is further explained as
follows.
Entity Extraction
This function is responsible for extracting entity
identity information from heterogeneous sources. We
assume that each entity, for example, an organization,
or a person, has a name, which the name can be
formally registered name or a known/nick name.
With this information, external data sources are
scanned in order to detect and extract identification
information related to this entity name.
Entity Reconciliation
This is the key function that forms the heart of the
EIDER platform. It reconciles variety of entity
identities, which are extracted from different sources,
and applies the intelligent algorithms for
reconciliation. The algorithms are comprehensive
methods that consist of string similarity metrics,
string distance function, natural language processing,
text mining, and graph traversal (Maali et Al, 2011).
Data Integration
After entity identities are reconciled, more
information can be extracted and integrated by using
the reconciled identities. A very important usage of
this function is to build reports, which the user might
want to extract, for example, balance sheet, or
publicly available finance performance data and
integrate them into a single report. This allows users
to gather more comprehensive knowledge from the
Big and Open Data to enhance their decision-making
(Insights on Data Integration Methodologies, 2008),
Figure 2: EIDER System Architecture.
ICEIS2015-17thInternationalConferenceonEnterpriseInformationSystems
406
although we do not restrict the application on private
data as well.
Apart from the key functionalities that the EIDER
system provides, another important aspect that worth
mentioning here is the data freshness. Since
information extracted by the EIDER system are Open
Data, freshness is crucial in order to maintain the
quality of data. EIDER employs patented method
(EP14176955.4) to retrieve data at system runtime
thus guarantees information in the EIDER system is
as fresh as possible.
4.2 Virtual Graphs and EIDER Model
We see two layers of data reconciliations in the
underlying system:
EIDER System initial entity reconciliation
Virtual Graphs enhanced entity reconciliation
The initial entity reconciliation is mainly text
mining technology focused, which the main purpose
is to allow accurate external entity identification
extraction from multiple Open Data sources. It is a
pre-reconciliation that makes the Big and Open Data
ready for integration.
At the end of EIDER system processing chain,
Big and Open Data are transformed into a common
analyzable format that is Linked Data. These data are
then stored into storage as graphs. To further
strengthen the entity reconciliation function of the
EIDER system, here we employ the Virtual Graphs
technology as post-reconciliation that can apply
graph algorithms to solve problems in the graph
space.
It is worth noting that we have conducted a
systematic literature review, and the results show that
currently, there are no any methodologies, tools, or
proposals that use Virtual Graphs for solving this kind
of problems.
To further explain our solution, we illustrate our
theory in the following figure:
Figure 3: Virtual Graphs and EIDER Model.
In this diagram, Big and Open data from
heterogeneous sources in different data formats are
extracted, (for example, DBpedia (Auer et al, 2007),
Yahoo Finance, GMEI Utility etc.); these data are
then integrated into a graph storage, whose variety of
entity identifications are managed by the Virtual
Graphs. Each node of the graph represents a unique
identifier from external data sources, and the edges
are the relationships between the different nodes.
With the dynamicity nature of the Virtual Graphs, we
are able to maintain an entity reconciliation system,
which is capable of adding/removing any new/old
entity identities at anytime, without breaking the
integrity of the whole data structure in the system.
5 CONCLUSIONS
Entity identification problem is a relatively new topic,
which is emerged with the Big and Open Data
phenomena. The only reference we have found so far
is the white paper presented by James and Nigel
(Powell and Shadbolt, 2014) in 2014. Nevertheless,
the focal point of that paper, is only to present
concerns to organizations and individuals, that they
should be cautious whenever they inventing a new
unique identifier within their system, for the entities
that are already exist.
Our paper addresses the entity identification issue
from a research and engineering perspective. With
our solution - a system that consists of an intelligent
reconciliation platform, the EIDER model, and the
Virtual Graphs technology, we are able to reconcile
multiple entity identification from heterogeneous
Open Data sources. Furthermore, the two layers of
reconciliation make sure the accuracy of the
reconciled entity identifiers. This is not only based on
historical data, but we are also able to incorporate any
new identities at dynamically at system runtime.
From a software engineering’s point of view, our
system is engineered in a generic manner, so that the
solution is applicable to many different domains that
share the same issue, as described in section III.
6 FUTURE WORKS
This paper presents an initial investigation of the
entity identification problem in the Big and Open
Data era. At the time of writing, we have completed
our initial system architecture design; some of the
core components in the EIDER system have also been
implemented, e.g. entity extraction, data integration,
and some basic entity reconciliation. For the forth-
EntityIdentificationProbleminBigandOpenData
407
coming period, we will focus more on enhancing the
intelligent reconciliation algorithms, in particular, the
Virtual Graphs technology for entity identification
and the integration into the EIDER system. A
schedule of tasks for this research work is show in the
figure 4.
The first step is to define a formal mechanism to
for the Virtual Graphs in the context of the Big and
Open Data. The second step is then to define a
methodological environment, which allows us to be
able to customize our approach to more concrete
scenarios, e.g. monument identification, e-Health,
and financial application.
Since the Virtual Graphs design relies completely
on practical experiences for each of the scenarios,
therefore, we need to work along these lines:
The definition of the methodological process for
the adaptation and application of unique
identification in Big and Open Data.
The definition of a formal procedure that will be
instanced for each concrete application scenario.
The definition of the procedure to carry out this
instance in a Virtual Graphs.
The practical evaluation of the approach in real
contexts.
Figure 4: Schedule of tasks.
ACKNOWLEDGEMENTS
This research has been supported by the MeGUS
project (TIN2013-46928-C3-3-R) of the Spanish
Ministry of Science and Innovation.
REFERENCES
Manyika, J., Chui, M., Brown, B. 2011. Big data: The next
frontier for innovation, competition, and productivity.
Official websites of government: data.gov and data.gov.uk.
Last check November 2014.
Berners-Lee. July 2006. Linked Data – Design Issues.
Berners-Lee. 2005. Uniform Resource Identifier (URI):
Generic Syntax.
Fielding, 1999. R. Hypertext Transfer Protocol.
Mondal, J., Deshpande. 2012. Managing large dynamic
graphs efficiently. A. pp. 145–156.
Powell, J., Shadbolt. S.N. March 2014. Creating Value with
Identifiers in an Open Data World. Open Data Institute
and Thomson Reuters.
Maali, F., Cyganiak, R., Peristeras. V. 2011. Entity
Reconciliation Against LOD Hubs.
Insights on Data Integration Methodologies. ESSnet-ISAD
workshop, Vienna, 29-30 May 2008.
Auer, S., Bizer, C., Kobilarov, G., Lehmann, J., Cyganiak,
R., Ives, Z. 2007. DBpedia: A Nucleus for a Web of
Open Data. pp 722-735.
ICEIS2015-17thInternationalConferenceonEnterpriseInformationSystems
408