An Ontology for Representing and Extracting Knowledge Starting
from Open Data of Public Administrations
Patrizia Agnello and Silvia Ansaldi
Department of Innovative Techonologies, Research Center, Inail, via Fontana Candida, 1,
00040 Monteporzio Catone, Rome, Italy
Keywords: Occupational Accidents, Open Data, Ontologies, Public Administration Knowledge.
Abstract: As proposed by European Commission through the institution of Europe’s Digital Agenda, the Italian
Digital Agenda has promoted the publication and the use of Open Data (OD) owned by the Public
Administration (PA), providing with the appropriate guidelines for describing and managing Open Data
under criteria of transparency, usability, accessibility. Inail (Italian Institute for Insurance against Accidents
at Work) has recently started to publish periodically OD related to occupational accidents and diseases. The
Inail OD, as they are defined, are mainly suitable for statistical studies but together with the vocabulary and
the typological tables, that explicitly describe some codes in the dataset, it has been possible to develop a
conceptual model which may be linked to schemas provided by other PA’s for achieving the interoperability
among those data through their semantics. Starting from the OD on occupational accidents, an ontology has
been developed for capturing the meaning (semantics) of information, both contained into the OD available
from Inail and defined in the vocabulary. The formalism of the ontology adopted for representing the
knowledge, in terms of concepts and their relations, ensures to share a common language avoiding
misinterpretations and misunderstanding.
1 INTRODUCTION
The challenge that we propose in this work is to
structure a set of ontologies starting from the “static”
vision of a phenomenon (accidents at work) that
provide open data. The basic idea is to establish a
robust platform, complete link for interoperability
with respect to shared data with other PA deriving
from open data set metadata.
In spite of other Italian PA, where ontology is
mainly adopted for presenting data in a quick and
trendy mode, linking them from different sites, the
original idea of our proposal is that the ontology
becomes a means of analyzing the “real” accidents
datasets, not only a means of presenting open data.
The goal is to allow, later, to assess cause and
effect, with the aim to prevent and, if possible,
reduce the high number of accidents themselves.
The choice of starting from the open data stems
from the fact that the information selected for the
open data describing the accident happened, and in
addition, some elements that determine the severity.
Are left out of the administrative, medical and
economic, being its internal aspects of the scope of
insurance, would contribute only to have a very
complex knowledge domain. Moreover, these latter
aspects would not contribute to the analysis of the
cases, however, for the prevention of injury.
The experience gained in the development of
ontologies comes as part of the industries and the
risk of a major accident. Developments have been
produced on the one hand for the management of
knowledge hidden in large quantities of documents
and the other as an aid in operational choices, for
example the most useful devices PPE (Agnello et al.,
2014) or more effective methods for the analysis of
the risk in industrial areas increasingly complex
(Ansaldi et al., 2012).
2 OPEN DATA
The Public Administrations, in order to perform
their tasks, gather, organize and manage a huge
amount of data, of different type and format. Most of
that data are public (by law), and therefore they may
be made open, that means available for others to use
(Dati.gov, 2015). These considerations are valid
195
Agnello P. and Ansaldi S..
An Ontology for Representing and Extracting Knowledge Starting from Open Data of Public Administrations.
DOI: 10.5220/0005557501950200
In Proceedings of 4th International Conference on Data Management Technologies and Applications (DATA-2015), pages 195-200
ISBN: 978-989-758-103-8
Copyright
c
2015 SCITEPRESS (Science and Technology Publications, Lda.)
both for national and European level (Reddick and
Anthopoulos, 2015).
The Open Data can create adding value in
various areas, e.g. to assure transparency of the
administrative actions, to share information among
the PA’s, or to reduce the gap of PA with citizens
(SPCData, 2015).
2.1 Inail OD
Inail has quite recently published on the owner site
the OD related to the occupational accidents and
occupational diseases, but only the first type has
been the objective of this research.
The OD are available in the standard formats (i.e.
RDF, XML, CSV), and can be downloaded free.
Each Inail OD record represents a case of an
accident, provided with a unique identification
number, and other 24 fields corresponding to
different types of information. The peculiarity of the
OD is the fine granularity, which means that
occupational accidents are individually represented
in separate records, without applying any rules for
combining them into some clusters.
The OD are delivered on monthly or half-yearly,
and are split per Italian regions. At the download of
the record set, a textual description provides with the
explanation of each field contained in the OD, i.e.
metadata name and its value type (OD Inail).
There exist two types of OD available, those
related to the occupational accident data and those,
which explain some codes, used in the OD.
Both of them are formally treated in a similar
way; indeed, they use the same format types and
textual description for the fields or metadata, but the
former correspond to the real data, while the latter
give the definitions, which are common to all OD.
Those data correspond to reference tables, each for a
specific topic, represented by a set of attributes.
Further to the textual explanations, enclosed with
the OD, a vocabulary (Ciriello et al., 2013) provides
detailed descriptions of all the terms used in the
context of insurance of occupational safety (at least
in the Inail terminology).
2.1.1 OD for Occupational Accidents
Inail decided to deliver open data in twofold aspects,
as individual information and as aggregate format.
As individual information, OD correspond to each
occupational accident; as aggregate, they are
organized into pre-defined clusters grouped by
specific topics. The table 1, in the first column,
shows the groups that characterize the types of
attributes expressed in the OD; in the second
column, some examples of the attributes are given.
With “Ref. Tables”, we indicate that the attributes
refer to one or more reference tables which are
properly of Inail “language”.
Figure 1: A diagram for the developed ontology.
DATA2015-4thInternationalConferenceonDataManagementTechnologiesandApplications
196
Table 1: Groups of attribute types and some examples.
Group Example
Temporal position of accident Dates
Geographical location of accident Province
Injured characteristics Nationality
Modality of accident At workplace
Administrative data of accident Ref. Tables
Medical features of accident Compensation
Owner characteristics Ref. Tables
2.1.2 Reference Tables
As previously described, some reference tables are
published using the same methods for representing
the OD. There are three types of reference tables:
Simple tables: which represent all the values that
an attribute may assume;
Complex tables: which contain multi-level
structures;
Tables with intervals: which are used to evaluate
a parameter with respect to given ranges and
determine the right classification of the injured
2.1.3 Statistical Tables
The OD, available for occupational accidents, are
expressed in a fine granularity, indeed, each record
corresponds to a single accident.
Furthermore, as previously discussed, some
tables, that group the data on different
characteristics, are also published on the site. They
reflect the statistical results, which are considered of
the major impact for the public interest.
For example, a set of tables groups the accidents
on the basis of the modality and period of
occurrence, others deal with the fatality accidents.
3 THE ONTOLOGICAL MODEL
The OD, as illustrated above, are well organized and
structured data, mainly useful for developing
statistical analysis, also sophisticated. Due to their
fine granularity, diverse aggregate types can be
studied, further to those statistical tables already
available on the site.
On the other hand, the high specificity of the
domain (insurance of occupational accidents) with
the particular vocabulary adopted can arise some
difficulties for the interoperability with other
contexts. Indeed, for the current version of OD, the
interoperability is represented only with the
information related to NACE code, for defining the
occupational sectors, the Italian Provinces, for
describing where the event happened, and a
classification of Nations for identifying the
nationality of the injured.
What we have done is to define an ontology
starting from the OD and the vocabulary, rather than
model the domain of “occupational accidents” in
general. The figure 1 shows the overview of all the
concepts modelled into the ontology and their
relations, some of them will be described in more
details in the paragraphs §3.3 and §3.4.
The reason for this choice is that the domain of
“occupational accidents” is characterized by sub-
domains of diverse nature, such as management,
legal and medical assessment, and economical
evaluation. To model this domain starting from each
single sub-domain in a complete form would be a
very complex project, requiring the efforts of
different types of knowledge and expertise. Since
the data available are only those represented in the
OD, it seems that the maintaining the consistency
between the sets of data available and the domain of
knowledge would be more effective.
Furthermore, since the granularity of the OD, the
data contained can be considered the main and the
sufficient information, which describes the event
(accident). Probably, in order to define the
administrative process for evaluating accident cases,
an ontology fully devoted for covering all the
administration aspects would be required.
All the fields contained into the dataset represent
a result of the evaluation process of the accident,
however, other information, which are considered
during the evaluation phase, are actually out of
scope of the OD, such as the description of
cause/effect related to each case. The defined
ontology can be easily extended to model those new
concepts; this approach seems to be more feasible
rather than trying to describe a general ontology,
which covers all the aspects.
3.1 Definition Process
The design and development of an ontology may
follow different methodological approaches (Noy
and McGuinnes, 2001). In this research, both
bottom-up and top-down approaches have been
adopted, combining the two methods in an iterative
process.
Indeed, starting from the data (contained into the
OD and the reference tables) and the definitions
enclosed into the vocabulary, through a bottom-up
approach, it was pointed out the concepts and the
instances to be modelled.
The top-down approach has been adopted for
AnOntologyforRepresentingandExtractingKnowledgeStartingfromOpenDataofPublicAdministrations
197
getting the overview of the domain. Indeed, three
elements are the key concepts to which all the other
data refer. They are the case of accident, the injured,
and the owner. The first two information are
necessary for the domain, otherwise data are not
meaningful. An identification number also uniquely
represents these three elements in the OD.
3.2 Architecture
The methodology followed to define the ontology is
to define many ontologies, each covers a domain,
even if it is small, but it has its own consistency and
conceptual autonomy of existence. This
methodology seems to have advantages both in order
to the extendibility of the ontology, previously
mentioned, and in the concept of its reusability.
Figure 2: Architecture of ontoInfortuni.
In figure 2, the architecture of the ontologies
developed is depicted. The central box indicates the
accidents ontology (ontoInfortuni), and the other
those representing the reference tables described in
§2.1.2. The different colours correspond to different
PA who are in charge of those types of data. The
upper boxes refer to data of property of Inail, while,
for example, NACE code is of Istat (Italian National
Institute of Statistics).
The box Administrative data Types refers to
concepts used for classifying injuries from the
administrative point of view, while Medical data
Types refers to models adopted for evaluating the
severity degrees of injuries.
The following paragraphs illustrate the main
ontology (ontoInfortuni) and the subsidiary
ontologies developed. The ontology editor used for
their definition is Protégé.
3.3 ontoInfortuni
Three concepts characterize the main ontology
(ontoInfortuni): the case of accident, the injured, and
the owner. All of them correspond to classes of the
ontology.
In turn, the case of accident is subdivided into
two subclasses: the occupational accident and the
fatal accident. The concept “occupational accident”
indicates that the injury has been formally
recognized by the administration process, and can be
insured by the Institute.
The “fatal accident” is itself subdivided into two
disjoint subclasses, depending on the outcome of the
insurance inquest, with positive or negative types.
There is a reference table containing all the codes
adopted for defining the outcomes of inquest of fatal
events (DecisioneIstruttoriaEsitoMortale).
Therefore, each subclass is defined as equivalent
to a class, which has relations with positive or
negative outcomes, that means a partition of the
codes in the reference table. The figure 3 shows a
graphical representation where the orange balls
correspond to the subclasses; the boxes represent the
individuals of the reference tables. The diverse
colours used to represent the individuals (dark for
background and light for foreground, or vice versa)
is only a graphical expedient for underlining that
each code correspond to a specific type of outcome,
positive or negative.
All the three classes have relations with the
appropriate reference tables or with simple data type
values (e.g. integer, string, date), which reflect the
attributes present in the OD. Furthermore, the
reasoner tools would be able to classify
automatically each case of fatal accident, as positive
or negative depending on their associated codes.
Nevertheless, in the OD each record corresponds
to an accident report, all data are at the same level in
sort of a flat representation. The identification
numbers (corresponding to the injured and the
owner) are unique for each individual but their
values are repeated in case of multiplicity of the
occurrence.
The ontology developed takes advantages of the
OD representation, further to the vocabulary,
however, the approach adopted is to model the
concepts enclosed into the OD’s, not their records.
So far, each identification number, through a
necessary and sufficient condition, uniquely
characterizes and classifies each individual of a class
(accident, injured or owner).
Each record of the OD implies that those three
main concepts, and so far their classes, are related to
each other. Two relations are explicitly modelled;
both of them have the case of accident as a domain.
One relation (refersTo) makes the event in
correspondence with owner (range); the other
DATA2015-4thInternationalConferenceonDataManagementTechnologiesandApplications
198
(occurrsTo) relates it with the injured (range).
This solution offers some advantages, further to
avoid repeating the same individuals, but also to
query the ontology.
3.4 Modelling Reference Tables
As discussed in the paragraph §2.1.2, some
reference tables contain the codes, which correspond
to pre-defined elements. Those tables have various
meaning, diverse proprietary and different structure,
as discussed in §3.2.
To represent the tables the same approach used
for the main ontology is adopted that means to
privilege the model of the concept rather than the
record of the table. In practice, the simple tables,
that contains a list of codes and their descriptions are
modelled as individuals belonging to a class with
two data type properties, the code and its
description.
The other more complex tables, e.g. NACE
classifications or Italian provinces, the approach is
to model them starting from the concepts they
describe; such as the hierarchical levels for NACE
representation, or the provinces, regions and macro
regions for the second type of table . However, those
models represent only an exercise of modelling,
since they are officially represented as ontologies
available in the public domain.
The table, whose codes are not absolute values,
but vary on a range of data, represents the last type.
The class deals with the disability degree (p) of the
injured. The table 2 shows, in the first column, the
intervals assumed by p, in the second column, a
description of type of disability.
Table 2: Disability table.
Range (p) (%) Disability
[1 - 5] Micro permanent
[6 - 15] Minimal but superior than micro
[16 – 25] Lower medium
[26 - 50] Upper medium
[51 - 85] Macro permanent
[86 - 100] Loss of health
It is important to note that the OD contains only
the value p, so far the classification of the events
with respect to the table would be a process
developed separately.
The model of this table is a class with six disjoint
subclasses; each of them is equivalent (necessary
and sufficient condition) to a class where the
parameter p (the disability degree) belongs to the
corresponding range. In addition one more subclass
for accidents without disability.
Even in this case, any reasoner tool is able to
classify automatically the events (accidents).
3.5 Browsing the Ontology
The system Protégé, used as ontology editor, has
interesting plug-in for querying and browsing the
model In our project, we used OntoGraf for
graphical representations of the classes, individuals
and relations; SPARQL Query for quickly
interrogate the ontology; OWLDoc functionality
Figure 3: Graphical representation of fatal accidents and the positive or negative outcome of inquest.
AnOntologyforRepresentingandExtractingKnowledgeStartingfromOpenDataofPublicAdministrations
199
which automatically generates pages in HTML for
browsing into the model. However, for better
represent and interrogate the ontology some
applications have been implemented. One
application is a graphical representation that tries to
show pictures which cannot be produced by using
OntoGraph plug-in, as illustrated in figure 3.
The other application is a prototype developed
for querying the ontology using SPARQL language.
As discussed by (Boselli et al., 2014), SPARQL is
powerful but requires a deep knowledge of both the
syntax of the language and the ontology model.
Both of these requirements are not common in
the users who could take advantages of the Open
Data. For this reason, this prototype provides the
user with a list of both predefined questions in a
natural language and statistical tables.
Each selection automatically generates an
appropriate SPARQL query, which can be processed
by the user. Furthermore, to write directly SPARQL
queries is always possible.
4 CONCLUSIONS
The proposed work is a different approach to open
data, at least for the Italian solutions. The ontology
presented does not want to be the instrument of data
presentation, as being those of fine granularity (each
record in the table regional accident corresponds to a
single injury) do not require display modes other
than standard, and there is no pre knowledge
developed by associate.
The OD Inail, related to accidents, publish a
selection of information extracted from a much more
complex and varied domain, covering different
aspects in the procedure following the accident of
the occurrence to liquidation economic.
The set of ontologies that are ontoInfortuni
model part of the information predisposing to linked
open data with other administrations, and a part of
knowledge based on the characteristics of the
accident itself will become the key to access to those
informative dataset, from where open data were
extracted for analysis of prevention. The latter is the
one on which you are currently working.
ACKNOWLEDGEMENTS
Our acknowledgements are for the authors G. Ciriello, M.
De Felice, R. Mosca, M. Veltroni of the Inail Research
Book n.1 since this ontology is based on the vocabulary of
Inail special language on occupational accidents. The
authors are also grateful to Prof. De Felice and Mrs.
Mosca for their support for the discussion and clarification
of specific topics related to occupational accident
insurance.
REFERENCES
Agnello P., Ansaldi S., Bragatto P.A., 2014. Pooling
knowledge and improving safety for contracted works
at a large industrial park. In Special Issue WORK: A
Journal of Prevention, Assessment, and
Rehabilitation. Karen Jacobs, Editor.
Ansaldi S., Monti M., Agnello P., Giannini F., 2012. An
ontology for the identification of the most Appropriate
Risk Management Methodology, P. Herrero et al.
(Eds.): OTM Workshops, Lecture Notes in Computer
Science 7567, 444–453. © Springer-Verlag Berlin
Heidelberg.
Boselli, R., Cesarini, M., Mercorio, F., & Mezzanzanica,
M., 2014. Are the Methodologies for Producing
Linked Open Data Feasible for Public
Administrations?. In Proceedings of Special Session in
Knowledge Discovery meets Information Systems
(KomIS) at DATA2014. ISBN 978-989-758-035-2
Ciriello, G., De Felice, M., Mosca, R., Veltroni, M., 2013.
Infortuni sul lavoro. Un modello di lettura (della
numerosità) su "open data" dell'Inail, Quaderni di
ricerca. Inail, Roma.
Noy, N. F., McGuinnes, D. L., 2001. Ontology Devel-
opment 101: A Guide to Creating Your First Ontol-
ogy. Stanford Knowledge Systems Laboratory Tech-
nical Report KSL-01-05 and Stanford Medical
Informatics Technical Report SMI-2001-0880.
Reddick, C. G., Anthopoulos, L., 2015. Information and
Communication Technologies in Public
Administration. Taylor and Francis Group. ISBN:
978-1-4822-3930-0
Protégé: http://protege.stanford.edu/
Agid: http://www.agid.gov.it/agenda-digitale
OD Inail: http://dati.inail.it/opendata/default/Infortuni/
index.html
SPCData: http://spcdata.digitpa.gov.it/index.html
Dati.gov: http://www.dati.gov.it/
DATA2015-4thInternationalConferenceonDataManagementTechnologiesandApplications
200
