=Paper=
{{Paper
|id=Vol-1551/article-06
|storemode=property
|title=None
|pdfUrl=https://ceur-ws.org/Vol-1551/article-06.pdf
|volume=Vol-1551
|dblpUrl=https://dblp.org/rec/conf/semweb/CottonG15
}}
==None==
https://ceur-ws.org/Vol-1551/article-06.pdf
Modeling the Statistical Process with Linked Metadata
Franck Cotton and Daniel W. Gillman
INSEE, Paris, France
franck.cotton@insee.fr
US Bureau of Labor Statistics, Washington, USA
Gillman.Daniel@bls.gov
Abstract. The official statistics community has recently put a lot of efforts into
the common formalization of its business semantics: information, process or
capability models, architectural patterns, etc. One of the first results of these ef-
forts, the Generic Statistical Business Process Model (GSBPM) is now a
worldwide reference for the official statisticians. But standard models, like all
metadata, need to be formally specified and machine-actionable in order to be
used in the most efficient way. This idea drives the preliminary work presented
in this paper, which aims at representing the GSBPM as an RDF vocabulary.
We explain the approach chosen to link the model to existing ontologies, detail
the results obtained so far and present some use cases and future developments.
Keywords: Statistical process, provenance, metadata, linked data, OWL.
1 Introduction
Metadata have always been essential for the statisticians. Structural metadata (codes,
concepts, data structure definitions…) specify how statistical products are constructed
and organized. Descriptive metadata document the methodology, the quality of the
products or how they may be used. Process metadata describe and measure the statis-
tical activities that transform raw data inputs into disseminated products.
In the recent years, the notion of active metadata has gained in importance. In this
paradigm, the metadata are not simply after-the-fact documentation of the statistical
data or treatments, but rather their "source code": they pilot the processes or generate
the statistical tools. This guarantees that the metadata are always in sync with the
actual operations. Illustrations of this approach can be found in [1] and [2].
Another trend observed recently in the statistical community is the growth of the
international collaboration. The UNECE1 is playing a key role in this phenomenon, in
particular through its High-Level Group for the Modernization of Statistical Produc-
tion and Services (HLG)2. A number of collaborative projects led by the HLG have
produced valuable results that are now international standards for the community, in
1
United Nations Economic Commission for Europe, http://www.unece.org.
2
http://www1.unece.org/stat/platform/display/hlgbas/High-
Level+Group+for+the+Modernisation+of+Statistical+Production+and+Services
particular a statistical information model3 or a reference business architecture for the
statistical industry4.
In this paper, we focus on another output of the HLG, the Generic Statistical Busi-
ness Process Model (GSBPM), which has recently become a worldwide de facto stan-
dard in the official statistics community. More and more statistical offices describe
and organize their business processes in reference to this model.
But the GSBPM is just a Word document at the moment: it is not expressed in a
formal and machine-readable form, and thus cannot easily be referenced or used in
the current efforts for building a globally shared active metadata corpus. The objec-
tive of this paper is precisely to propose a first attempt at building a simple linked
(meta)data representation of the model, so that it can offer the benefits of the linked
data paradigm (unique global identification, shared semantics, linkability, etc.) and
fully play its role as a common vocabulary and central reference for the statistical
community.
The paper is structured as follows: in Section 2, we describe in greater detail the
GSBPM and how it is organized. Section 3 gives a quick overview of previous works
and existing standards that can be leveraged for our purpose. The construction of the
GSBPM vocabulary itself is presented in Section 4, and Section 5 lists some intended
use cases. Finally, Section 6 provides a summary and conclusion, and discusses poss-
ible future work.
2 The statistical process
In some industries, there is a common core business process all producers follow.
This is true for official (national and international) statistics, where statistical data are
the products of the activities of statistical institutes, and each institute follows the
same general practice. Traditionally, statistical offices have used censuses (where
every person or business is contacted) or sample surveys (where a select few, or sam-
ple, of persons or businesses are contacted) to collect data. These common approaches
mean that all statistical offices enjoy similar successes and face the same set of prob-
lems. This, in turn, has led to a de facto standard way to conduct the business of offi-
cial statistics.
More recently, as sample surveys have become significantly more expensive to
conduct, statistical offices have looked at other sources of data, such as administrative
records and Big Data, to augment or replace the traditional methods. However, the
fundamental business process has not changed that much. Statistics and probability
are used in the underlying mathematical model for handling survey design, and an
advantage to this is that the model can be applied to data collections that are not sam-
ple surveys as well.
3
GSIM, Generic Statistical Information Model,
http://www1.unece.org/stat/platform/display/gsim/Generic+Statistical+Information+Model.
4
CSPA, Common Statistical Production Architecture,
http://www1.unece.org/stat/platform/display/CSPA/Common+Statistical+Production+Archit
ecture+Home.
Over time, many statistical offices wrote process models or standards for describ-
ing their business activities. These models look remarkably similar across offices, and
an effort was started under the UN Economic Commission for Europe (UNECE) to
build a standard model. The effort is known as the Generic Statistical Business
Process Model (GSBPM).
2.1 The GSBPM
The effort to build the GSBPM started around 2005 under the UNECE Statistical
Metadata Working (METIS) Group within its project known as the Common Metada-
ta Framework (CMF). By 2008, the first draft was completed, and GSBPM became
the major deliverable of Part C of the CMF. Revisions to GSBPM have been pro-
duced periodically since.
Around 2011, the High Level Group for Modernization of Statistics (HLG) was
formed, and METIS was replaced by the Modernization Committee for Standardiza-
tion (MCS) under HLG. Refinements to the GSBPM are being managed under the
MCS. The fifth version of GSBPM was released in late 2013.
GSBPM will be described in more detail in the next section, but it is built as a set
of major process steps with additional detail under each. The terminology used to
name and describe each step is generic, and the terms do not necessarily correspond to
the terminology used in any country. Further, the standard is written in English, so
most countries have to translate GSBPM to their own language to make it most useful
locally. However, many countries have adopted GSBPM, and it is being used in many
novel ways. One common use is to classify statistical processing software, and this
provides a means to manage and classify statistical software and find areas where
statistical offices are unnecessarily duplicating software development activities.
GSBPM has also spawned two other newer standards under HLG/MCS, and these
are the Generic Activity Model for Statistical Organizations (GAMSO) and the Ge-
neric Statistical Information Model (GSIM). GAMSO, first produced this year, ex-
tends GSBPM to include non-statistical activities common to all statistical offices.
These include many management activities such as needs for computing infrastruc-
ture. GSIM, first released in 2012, is an information model describing the statistical
information objects necessary for statistical activities. GSIM and GSBPM are duals of
each other in the sense that GSBPM names activities and GSIM names the inputs and
outputs of those activities. In this way, GSIM and GSBPM are intertwined.
The GSBPM is not really a business process model. The flow is not described,
even if there is a timeline broadly going right and down. It is centered on the naming
of the different statistical activities that constitute the statistical process. In fact, it is
rather a taxonomy.
2.2 Model description
GSBPM is called a business process model, but it really is a taxonomy of terms that
name the activities conducted by statistical offices. Typically, process models include
the flows among all the processes, but GSBPM does not go to that detail. In fact, the
statistical process is fluid enough, that the order of many processes is
is not that impor-
impo
tant. Sometimes, steps can be conducted in any order with the same results. Therefore,
the higher level taxonomy is the lowest level of detail the model can provide and still
be useful world-wide.
Figure 1 below contains the top two levels of the GSBPM.
Fig. 1. Generic Statistical Business Process Model
There are eight main process steps, which are labeled in the blue boxes at the tops
of each of the columns. The items below each of the blue boxes are the sub-
su -processes
under the main process labels. Provided across the top are two over-arching
over arching steps,
quality and metadata – as these are important activities that go along with every
process in the model. In other words, every process generates metadata needed to t
describe that process and describe its usage; and every process has a quality compo-
comp
nent attached to it.
The main processes,, or phases, are defined as follows:
• Specify Needs - the initial investigation and identification of what statistics are
needed and what is needed of the statistics
• Design - the development and design activities, and any associated practical re- r
search work needed to define the statistical outputs, concepts, methodologies, col-co
lection instruments, and operational processes
• Build - build and
nd test the production solution to the point where it is ready for use
in the "live" environment
• Collect - collects or gathers all necessary information (data and metadata), using
different collection modes (including extractions from statistical, administrative
administ
and other non-statistical
statistical registers and databases), and loads them into the appropri-
appropr
ate environment for further processing
• Process - the cleaning of data and their preparation for analysis
• Analyze - statistical outputs are produced, examined in detail and made ready for
dissemination
• Disseminate - manage the release of the statistical products to customers
• Evaluate - manages the evaluation of a specific instance of a statistical business
process
Under each main process are defined several sub-processes. These define specific
kinds of processes under each of the main headings. Further detail can be described,
but at this level those details are probably institute specific.
The main result of the GSBPM is a taxonomy of business production activities.
This is made possible in part by the acceptance of common practices across the offi-
cial statistical community world-wide.
3 Choosing base vocabularies
When designing an RDF vocabulary, it is always good practice to relate it to existing
ontologies, in order to leverage field-proven models. It helps to clearly specify the
semantics, facilitates the creation of links between models or entities that is at the
heart of the Linked Data paradigm, and allows the reuse of existing knowledge.
The GSBPM being a business process model, a first line of work consist in looking
towards existing standards in the field of business process modeling (BPM). But we
also noted that the GSBPM could be seen as a taxonomy of statistical activities, which
leads us to examine the domain of knowledge organization systems. Finally, we can
refer to the transformative character of the statistical process and associate that to
information provenance models.
3.1 Business process modeling standards
One of the most prominent standards in the BPM domain is the Business Process
Model Notation (BPMN), maintained by the OMG5.
BPMN is a diagramming standard: it defines a way to represent process objects
like activities, data objects, events or gateways (where activity flow paths are com-
bined or separated). The most recent version of the standard is version 2 [3], follow-
ing versions 1.0, 1.1 and 1.2.
There has been a lot of work realized in the area of semantic BPM (SBPM), in par-
ticular around BPMN. For example, [4] presents a very comprehensive work made on
the semantic formalization of BPMN 1.1. There is also an ongoing work by the same
team for building a BPMN 2.0 ontology, but it is not achieved yet, to the best of our
knowledge.
The authors clearly indicate that the BPMN ontology models graphical symbols,
and not the represented process components. For example, all main classes extend a
"graphical-element" class. This is not the notion that we are looking for.
5
Object Management Group, http://www.omg.org.
In addition, the conformity to the BPMN model leads to introduce some constraints
that are not suited for our purpose (for example, the "activity" class must have a type,
start and completion quantities, etc.).
Another important BPM standard is WS-BPEL [5], or BPEL in short. Here also,
some SBPM references exist, for example [6]. This BPEL ontology introduces an
"Activity" class at the base of a hierarchy of more specific activities. There again, in
conformity with the objective of the ontology, what is modeled is the BPEL activity
(basically a combination of web services): the Activity is described by "Concept of
being a BPEL activity", whereas GSBPM phases or sub-processes, or even statistical
production activities, are of course not all activities in the sense of BPEL. Moreover,
the BPEL ontology does not seem to be available as OWL.
As a conclusion, both the BPMN and BPEL ontologies can be very useful in the
representation or annotation of the detailed statistical processes that the GSBPM is
meant to classify, but neither provides the notion of generic statistical activity (or
even simply activity) that we would need here. The authors of [4] refer to the
DOLCE6 upper ontology for the modeling of the processes themselves, so we could
envision to use it also, or its lighter OWL manifestation DOLCE+DnS Ultralite
(DUL)7, but the Process as defined in DOLCE is a very generic concept, and DOLCE
is based on an ontological approach, very well described in [7], that needs to be stu-
died further to see if it is appropriate to our use case.
3.2 knowledge organization systems
We saw in the introduction that GSBPM activities and sub-processes are rather ac-
tivity categories where actual statistical activities can be classified. In that sense, the
model is close to a taxonomy of statistical activities, and one can think of describing it
as a concept scheme as defined in the Simple Knowledge Organization System
(SKOS) [8], by far the most used knowledge organization vocabulary in the RDF
world.
In such a representation, the GSBPM phases and sub-processes, as well as the
whole statistical production process itself, would be viewed as concepts belonging to
the GSBPM concept scheme. In SKOS terms, a sub-process would have for broader
concept the phase to which it belongs, this phase having itself the whole statistical
process as broader concept.
Using SKOS also has the advantage of being able to leverage the different docu-
mentation properties defined in the recommendation (skos:note and all its sub-
properties), in order to store the abundant textual material contained in the GSBPM
specification. It also directly enables the use SKOS extensions, like the XKOS voca-
bulary [9] which is already known in the statistical community and which defines in
particular semantic relations between concepts that refine those defined in SKOS, for
example partitive or generic/specific relations.
6
https://en.wikipedia.org/wiki/Upper_ontology#DOLCE_and_DnS.
7
http://ontologydesignpatterns.org/wiki/Ontology:DOLCE%2BDnS_Ultralite
3.3 Provenance
rovenance models
cabularies (LOV)8,
When searching for the term 'activity' in the Linked Open Vocabularies
the first answer returned belongs to the PROV ontology, or PROV-O [10].
PROV is a set of W3C Recommendations that defines a data model and various se- s
rializations to support the representation and exchange of provenance information.
information
It is based on a very simple and elegant core model:
Fig. 2. PROV Core Structures (copyright © 2011-2013 W3C®)
In this model, an entity is "a physical, digital, conceptual, or other kind of thing
with some fixed aspects", and an activity is "something that occurs over a period of
time and acts upon or with entities". These definitions are quite generic, but the model
represented above gives them a semantic context that seems immediately relevant to
the statistician. Statistical
Statistic production is all about informational entities that derive
from each other through a succession of activities informed by one another.
Looking more precisely into the complete PROV model, it is easy to verify that it
contains all the information needed to implement process traceability. This is a very
important feature for the statistical organizations, which put more and more efforts
into quality insurance.
3.4 Conclusion
To conclude this section, we see that the specialized ontologies defined in the
SBPM fieldld in relation to well-known
well known standards like BPEL or BPMN are too specia-
speci
lized to be used in the representation of the GSBPM, but could be very useful for the
8
http://lov.okfn.org
modeling or annotation of more fine-grained activities that would be classified ac-
cording to the GSBPM.
In contrast, more high-level vocabularies like PROV and SKOS can easily be re-
lated to the model, and they bring valuable additional semantics that can add to the
expressivity and usefulness of the model.
4 Building the vocabulary
This section builds on the previous ones and explains concretely how we design
the GSBPM vocabulary. Given that PROV-O and SKOS/XKOS are expressed in
OWL, we adopt the same formalism, although in practice only RDFS statements are
used at this early stage, except for the ontology object itself (a owl:Ontology).
The code snippets9 given below as examples use Turtle for the representation of RDF
1.1 statements, and the usual prefixes for the well-known vocabularies as documented
in http://prefix.cc/.
For now, the vocabulary defines only classes and individuals: datatype or object
properties could be introduced in future versions.
4.1 Namespaces
The GSBPM is published and maintained under the auspices of the UNECE, so we
tentatively use as base namespace http://rdf.unece.org/models/gsbpm#
for the ontology, and http://id.unece.org/models/gsbpm/ for the indi-
viduals. Using two different namespaces emphasizes the special importance of global
identification for the GSBPM individuals. This design choice can easily be changed
and the two namespaces merged in the future. In any case, the namespaces must be
discussed and validated with the UNECE and HLG, in particular to implement derefe-
renceability of the URIs.
The prefixes associated with the namespaces mentioned above will be respectively
gsbpm and igsbpm.
4.2 Classes
As we saw in the description of the GSBPM, the central notion is "statistical pro-
duction activity". In order to capture this conclusion, we create a base class named
StatisticalProductionActivity. As a consequence of the study made in
the previous section, this class will be defined as a sub-class of both
prov:Activity and skos:Concept:
gsbpm:StatisticalProductionActivity
a rdfs:Class, owl:Class ;
rdfs:label "Statistical production activity"@en ;
9
The complete ontology is available at:
https://github.com/FranckCo/Stamina/raw/master/doc/gsbpm.ttl.zip
rdfs:subClassOf prov:Activity , skos:Concept .
This class is further specialized into gsbpm:Phase and gsbpm:SubProcess.
The aggregation relation between phases and sub-processes is not represented at this
level (it could be in OWL), but rather at the level of the individuals.
4.3 Individuals
The GSPBM model, considered as a global taxonomy of statistical activities, is
represented in the model as a skos:ConceptScheme (only a few properties are
reproduced below). This materializes a clear distinction between the model itself and
the statistical process which is modeled.
gsbpm:GSBPM
a skos:ConceptScheme ;
rdfs:label "Generic Statistical Business Process
Model"@en ;
foaf:homepage .
The statistical process as a whole is represented as an instance of Statistical-
ProductionActivity:
igsbpm:StatisticalProductionProcess
a gsbpm:StatisticalProductionActivity ;
rdfs:label "Statistical Production Process"@en ;
skos:topConceptOf igsbpm:GSBPM ;
skos:narrower igsbpm: 1 , igsbpm: 2 , igsbpm: 3,
igsbpm:4 , igsbpm:5 , igsbpm:6 , igsbpm:7 , igsbpm:8 .
In the same manner, all the phases and sub-processes defined in the GSBPM are
also instances of StatisticalProductionActivity. We give below example
definitions of a phase and a sub-process (only a few properties are shown, and texts
have been shortened for brevity):
igsbpm:4 a gsbpm:Phase ;
skos:notation "4" ;
skos:prefLabel "Collect"@en ;
skos:definition "This phase collects..."@en ;
skos:inScheme igsbpm:gsbpm ;
skos:narrower igsbpm:4.1 , igsbpm:4.2 , igsbpm:4.3
igsbpm:4.4 ;
skos:broader igsbpm:StatisticalProductionProcess ;
xkos:isPartOf igsbpm:StatisticalProductionProcess .
igsbpm:6.2 a gspbm:SubProcess ;
skos:broader gsbpm:6 ;
xkos:isPartOf gsbpm:6 ;
skos:definition "This sub-process is..."@en ;
skos:inScheme gsbpm:gsbpm ;
skos:notation "6.2" ;
skos:prefLabel "Validate outputs"@en .
The skos:broader and skos:narrower properties are used to formalize the
hierarchical structure of the GSBPM, doubled with partitive relations like
xkos:isPartOf for the use cases that understand XKOS.
We chose not to express relations of causality, sequencing or temporality between
phases or sub-processes because, as explained several times in the GSBPM textual
material, they can overlap or be altogether missing in some cases. In the description
of more precise processes, statistical activities could be connected by these kind of
relations using XKOS associative properties like xkos:causal,
xkos:sequential, xkos:disjoint or their sub-properties.
The reader may have notices that the over-arching processes of the GSBPM have
not been included in the model. This is intentional, because quality and metadata
management need to be considered in a broader perspective that is now covered by
the GAMSO. The articulation between the GSBPM and the GAMSO still need a bit
of work, and semantic formalization of both models will undeniably help in that re-
spect.
5 Use cases
As we indicated in the introduction, the formalization of the semantics of the GSBPM
is an objective in itself because the model is a reference framework for official statis-
tics. Representing it as a simple, yet precise, RDF model opens the possibility to dis-
cuss within the community in order to improve this representation and come to an
agreement on a shared semantics.
The use of RDF also provides a global naming scheme, so that we can refer unam-
biguously to the model or its specific components. Also, RDF being multilingual by
design, we can enrich the GSBPM vocabulary with translations of the textual descrip-
tions that it contains. Spanish and French translations of the previous version of the
GSBPM are available and could easily be upgraded and integrated.
Beyond this "intrinsic" utility, the work initiated here can be leveraged in different
use cases. We give below a few examples.
First, the vocabulary provides an anchor for national refinements of the GSBPM.
Several countries have developed more precise models where the sub-processes are
broken down into more precise activity descriptions. The approach exposed in this
paper can be reused to model these extensions and to attach them to the main model.
This will in particular allow for comparisons between the different variations that can
exist at national level.
Similarly, the GSBPM ontology can be used to organize and share formal descrip-
tions of actual statistical processes. If a country models a process, either with the
SBPM tools and techniques that we briefly reviewed in section 3 or by means of any
other formal representation, it can refer by URI to (and be referred by) the relevant
GSBPM sub-process. We see here the possibility to build a global reference of statis-
tical process descriptions, classified according to the GSBPM, that would be fully in
line with the CSPA approach (see below).
Another interesting use case has already been mentioned and is offered by the arti-
culation with the PROV ontology. PROV's very rich model opens great perspectives
for the documentation of the statistical process, which is basically a very complex
transformational process. The possibility to capture provenance information in a stan-
dard way along the whole flow, even at a relatively coarse-grain level, is a very inter-
esting one.
Finally, the simple fact to actually store the model in a RDF database with basic
capabilities like querying, full-text search or simple visualization would greatly im-
prove the dissemination of the GSBPM.
6 Conclusion and future work
In this paper, we presented a preliminary work on representing the Generic Statistical
Business Process Model (GSBPM) as an RDF vocabulary. We studied different stan-
dards existing in the field of Semantic Business Process Modeling like the BPMN and
BPEL ontologies, and concluded that they could be useful for representing actual
"real-life" statistical processes, but that the GSBPM itself was too high-level to be
directly related to them. On the contrary, more generic models like those of SKOS
and PROV can easily be used in that respect, and they bring valuable enrichments and
generate new use cases.
The work presented here is only a first step. The big test, of course, will be whether
the statistical community wants to use the GSBPM vocabulary. For that, we need to
consolidate the vocabulary, have the model validated by the community, make the
URI scheme and license model official, etc. We also should implement a quick and
easy use case like the basic visualization application mentioned at the end of Section
5. It would also be important to study in more detail the work done in the SBPM field
in order to provide advice to the statistical community on modeling processes and
attaching them to the GSBPM.
The work ahead must be put in the perspective of the international collaborations
going on. The Common Statistical Production Architecture (CSPA) is an effort to
standardize statistical production software across all statistical offices. This effort
involves the idea that production systems are built up from small modules that every-
one shares. However, GSBPM will guide the development of which modules must be
built, and GSIM will guide which metadata are inputs and outputs to each process.
GAMSO will extend the description to activities that are outside of statistics proper.
As these standards are formalized as linked metadata, it will be possible to move to-
wards the ultimate objective of a completely formalized semantics of the statistical
domain, in order to be in the position to implement the "Active meta-data vision".
References
1. Rivera, A., Wall, S. and Glasson, M.: Metadata-driven business process in the Australian
Bureau of Statistics (2013), downloaded from the web on 9 July 2015 at
http://www.unece.org/fileadmin/DAM/stats/documents/ece/ces/ge.40/2013/WP21.pdf.
2. Sigaud, É., Tailhurat, R., Cotton, F. and van der Vlist, É.: XForms generation, a real world
example (2013), downloaded from the web on 9 July 2015 at
http://www.balisage.net/Proceedings/vol11/html/Cotton01/BalisageVol11-Cotton01.html.
3. OMG, Business Process Model and Notation (BPMN), Version 2.0 (2011), specification
downloaded from the web on 9 July 2015 at http://www.omg.org/spec/BPMN/2.0/PDF.
4. Rospocher, M. Ghidini, C. and Serafini, L.: An Ontology for the Business Process Model-
ling Notation (2014), downloaded from the web on 9 July 2015 at https://dkm-
static.fbk.eu/people/rospocher/files/pubs/2014foisbpmn.pdf.
5. OASIS, Web Services Business Process Execution Language v2.0 (2007), specification
downloaded from the web on 9 July 2015 at https://www.oasis-
open.org/standards#wsbpelv2.0.
6. Nitzsche, J., Wutke, D., and van Lessen, T.: An Ontology for Executable Business
Processes (2007), downloaded from the web on 9 July 2015 at http://ceur-ws.org/Vol-
251/paper8.pdf.
7. Gangemi, A., Guarino, N., Masolo, C., Oltramari, A., Schneider, L.: Sweetening Ontolo-
gies with DOLCE (2002), downloaded from the web on 9 July 2015 at
http://www.researchgate.net/profile/Nicola_Guarino/publication/221630979_Sweetening_
Ontologies_with_DOLCE/links/0046352ce74538d87a000000.pdf
8. W3C, SKOS Simple Knowledge Organization System Reference (2009), recommendation
downloaded from the web on 9 July 2015 at http://www.w3.org/TR/skos-reference/.
9. Data Documentation Initiative (DDI) Alliance (2012), XKOS - Extended Knowledge Or-
ganization System, specification downloaded from the web on 9 July 2015 at
http://www.ddialliance.org/Specification/RDF/XKOS.
10. W3C, PROV-O: The PROV Ontology (2013), recommendation downloaded from the web
on 9 July 2015 at http://www.w3.org/TR/prov-o/.