MAINTAINING PROPERTY LIBRARIES IN PRODUCT
CLASSIFICATION SCHEMES
Joerg Leukel
Institute for Computer Science and Business Information Systems, University of Duisburg-Essen, Essen, Germany
Keywords: B2B e-commerce, product data, product ontologies, standardization.
Abstract: Semantic interoperability in B2B e-commerce can be achieved by committing to a product ontology that
establishes a shared and common understanding of a product domain. This issue is mainly subject of
standard product classification schemes. Recently, considerable research and industry work has been carried
out on enhancing the semantic richness of these schemes. Providing specific property lists for each product
class can be seen as a step towards true product ontologies. Horizontal classification schemes, however,
often consist of more than 10,000 classes, several thousand properties, and an even greater number of class-
property relations. Given the new requirement towards property-centric classification, maintaining these
business vocabularies is mainly determined by strategies for managing the property definitions and their
relationships to classes. This paper proposes measures for coping with the problem of extensive and steadily
growing property libraries. It can be shown that implementing these measures greatly influences both
standards makers and standards adopters.
1 INTRODUCTION
Executing business processes between independent
organizations faces often heterogeneity concerning
process models, data sources, software systems, and
available meta data describing these components.
Automating such processes increases the need for
aligning heterogeneities and finding consensus about
common concepts. Ontologies aim at fulfilling this
role by establishing a shared and common
understanding of a domain. In B2B e-commerce,
most processes incorporate essential information
about products (and services) being the subject of
procurement and sales respectively. Therefore, the
development of product ontologies can be regarded
as an enabler of machine-readable, unambiguous
representations of information about products
(Fensel et al., 2001) (Leukel, 2004).
A number of industry consortia and standards
bodies have proposed such domain ontologies called
standard product classification schemes (standard
PCS). Applying these business vocabularies benefits
searching for products in e-catalogs, comparing
similar products, standardizing product descriptions,
and facilitates spend analysis and product-sensitive
workflows (CEN/ISSS, 2005). Prominent horizontal
standards, such as eCl@ss, eOTD and UNSPSC
consist of 20,000 up to 60,000 product classes, and
represent a huge amount of knowledge about the
categorization of products.
For standards makers, the broad coverage of
horizontal standards leads to an enormous amount of
properties as new subjects of standardization
including proposal, negotiation, definition, and
maintenance. Taking in mind the distributed nature
of many standardization processes (e.g., work
groups for each sub-domain or branch of industry),
reducing or avoiding redundant properties becomes
an important task. For instance, work groups should
always check carefully the appropriateness of
existing properties before proposing a new property
for the vocabulary. This basic principle does not
affect the underlying model, but the standardization
process. It requires, however, that properties are
reusable. Reusability of properties depends on their
semantic precision, naming issues (i.e. synonyms,
homonyms), and the conflict between wide or
narrow definitions.
For standards adopters, properties must be seen
from a different perspective. Classification based on
a standard PCS requires (1) assigning each product
to a class and (2) describing each product with class-
specific properties. This initial effort is time-
3
Leukel J. (2006).
MAINTAINING PROPERTY LIBRARIES IN PRODUCT CLASSIFICATION SCHEMES.
In Proceedings of the Eighth International Conference on Enterprise Information Systems - SAIC, pages 3-10
DOI: 10.5220/0002442600030010
Copyright
c
SciTePress
consuming and costing; it depends mainly on the
number of products and the number of properties per
class. In addition, it requires continuous efforts
triggered by new PCS versions (new product classes,
new properties, redefinition of classes and
properties).
This paper proposes measures for coping with
the problem of extensive and steadily growing
property libraries. Our contribution to research is
that we (i) introduce the still overseen problem of
property growth, (ii) provide a comprehensive set of
measures that address this problem, and (iii)
reconstruct the current state of standard PCS by
checking whether some of the measures are
implemented as of today.
The remainder of our paper is structured as
follows. Section 2 discusses related work, and shows
that extensive property libraries have rarely been the
subject of research. In Section 3, we describe the
main problem based on empirical observations. In
addition, we present the basic conceptual data model
of property-centric classification schemes, which
will be extended in the course of our work. In
Section 4, we define measures for coping with
property libraries, and describe their impact on the
problem. In Section 5, we discuss the measures by
summarizing the impacts, extending the basic
model, and reviewing selected standard PCS.
Finally, we draw conclusions and point to future
avenues of research (Section 6).
2 RELATED WORK
Related work to product classification schemes can
be found in several fields such as e-catalogs, product
data management, standardization, and ontology
engineering. Next, we provide an overview of
closely related work and outline their relevance to
the problems caused by extensive property libraries.
Early work on basic concepts of PCS presents
and evaluates standard PCS from a business
perspective. For instance, (Fairchild & Vuyst, 2002)
examined the role of standard PCS towards benefits
of spend analysis; properties are not necessary for
this business function.
Schulten et al. introduced product classification
as a reference domain for ontology engineering and
the Semantic Web, and called for concentrated
efforts to “design a generic model” for automated
mapping between two different PCS (Schulten et al.,
2001). Concerning product properties, the proposal
demonstrated the mapping problem between classes
only, but did not incorporate properties. Eventually,
the research prototypes in (Corcho & Gómez-Pérez,
2001) and (Beneventano et al., 2004) followed this
class-centered path.
Ng et al. described challenges in integrating
product schemes based on heterogeneous properties
(Ng, Yan & Lim, 2000). Property lists form schemas
that can be integrated by applying techniques from
database schema integration. Two interesting aspects
discussed by Ng et al. are shallowness (flat
structures, lists) and bushiness (clusters of a high
number of related properties). This database
approach is complemented and extended in (Bullig,
Schnadhorst & Wilkes, 2003), which analyzed
property mappings and practical issues in more
detail.
Leukel et al. proposed an XML-based exchange
format for PCS (Leukel, Schmitz & Dorloff, 2002).
Its contribution lies in identifying and defining data
elements and relationships, both being derived from
an empirical study of four standard PCS and three
XML e-business standards. The modeling of
properties is quite sophisticated and fulfills mainly
requirements of PCS adopters. Two measures for
supporting the “management and maintenance” of
property libraries – grouping and inheritance – are
described briefly.
Recently, the importance of properties as a
cornerstone of product classification has become
more evident. For instance, (Ondracek & Sander,
2003) argued on a “property based product
classification” from that multiple different
classification hierarchies for specific purposes can
be built, though they are based on common, thus
standardized properties. (Leukel, 2004) emphasized
the role of properties for providing additional
semantics to class hierarchies; properties are needed
to describe the scope of a class formally. (Kim et al.,
2004) developed a “semantic classification model”
based on properties in order to enable an in-depth
understanding of product classification. All this
work is in support of semantically rich PCS that
incorporate well-defined properties. The problems
caused by large property libraries are being overseen
though.
A first indication of problems related to
properties can be found in (Hepp, 2004). The
proposed quantitative measurements for PCS reveal
some shortcomings in property lists and can help
detect duplicate properties. In its conclusion, the
paper argues on the need for further work on
maintaining properties and organizing property
libraries. A comprehensive quantitative analysis of
classes and properties in PCS is subject of (Hepp,
Leukel & Schmitz, 2005). In addition, (Ondracek &
ICEIS 2006 - SOFTWARE AGENTS AND INTERNET COMPUTING
4
Sander, 2003) drew attention to the problem of
redundant properties in huge property libraries, and
claimed that separating definition and application of
properties would be “the only solution”.
3 PROPERTY-CENTRIC
PRODUCT CLASSIFICATION
In recent years, considerable industry efforts have
been undertaken to extend the semantic richness of
product classes by adding class-specific property
lists. A property list contains all properties that
should be used to describe products belonging to the
respective class. These lists greatly enhance the
formal precision of standard PCS by replacing class
labels with a structured, though human-language
description of its meaning. From an ontology
perspective, property lists can be regarded as a first
step towards true product ontologies, since they
provide standardized representations of product
concepts, thus machine-readable semantics (Hepp,
2005). Taking in mind the already huge number of
product classes, adding property lists is not only a
resource-intensive project, but it may also cause new
problems for both standards makers and standards
adopters. We refer to these problems as of property
growth.
3.1 Observations
Property growth can be assessed by comparing
multiple versions of the same standard PCS. Next,
we present some data for eCl@ss, a horizontal PCS
being developed by a consortium of mainly German
companies since the late 1990s (eCl@ss e. V.,
2004). It has gained a significant relevance for e-
procurement in many European countries. Table 1
shows basic figures for four versions of eCl@ss.
The number of classes and class-property
relations has tremendously increased (107% and
492%). The increase of properties, however, is not
constantly over time (significant decline in version
5.0). We can assume that eCl@ss has already
performed some actions to limit redundancy by
reorganizing the property library, although, the sheer
number of property remains high (140% increase in
version 5.1 compared to version 4.0).
Table 1: Classes and Properties in eCl@ss.
eCl@ss
4.0
eCl@ss
4.1
eCl@ss
5.0
eCl@ss
5.1
Publication
Date
Aug
2001
Sep
2002
Dec
2003
Sep
2004
Classes
(4th Level)
10,190 12,565 20,379 21,100
Properties 2,303 5,504 3,667 5,525
Class-Prop.
Relations
68,244 303,511 406,482 403,859
Looking closer at eCl@ss (table 2), we observe
that (i) property lists are added to more and more
classes, (ii) most property lists contain at least 30
properties (V5.0: 83% of all property lists; V4.1:
maximum of 294 properties!), and (iii) the recent
version has led to a decrease in number of properties
per class as indicated by the mean.
Table 2: Property Lists in eCl@ss.
eCl@ss
4.0
eCl@ss
4.1
eCl@ss
5.0
eCl@ss
5.1
Classes with
Property Lists
1,107
(10.9%)
6,507
(51.8%)
7,913
(38.8%)
10,930
(51.8%)
Properties per
Class: Minimum
66 3 1
Properties per
Class: Maximum
89 294 266 156
Properties per
Class: Mean
20.6 42.0 43.5 32.3
Properties per
Class: Derivation
10.3 16.7 13.3 15.2
The growing number of property lists and
properties per class causes significant classification
costs, thus calls for property-centric classification
strategies for PCS adopters (i.e. suppliers, buyers,
marketplaces). Moreover, we have to consider that
in most industries multiple standard PCS are
relevant, especially due to the competition between
major horizontal standards. Each standard PCS
defines its own classes, properties and property lists.
Therefore, the problem of property growth is
multiplied by the number of standard PCS.
3.2 Basic Model
Product classes are the core components of each
PCS. A product class is a categorization, collection
or type of similar products that share a set of
characteristics (e.g., the class “laptop” describes
portable computers). Product characteristics are
expressed by properties (e.g., CPU type, display
MAINTAINING PROPERTY LIBRARIES IN PRODUCT CLASSIFICATION SCHEMES
5
size, weight). Properties are not limited to a single
class, but should be reusable. While some properties
represent information that can be captured by
standard data types (e.g., string, integer, float,
Boolean), other properties limit the allowed values
to a specific list of values (e.g., color “red”, “green”,
“blue” etc.). In conceptual modeling, the definition
of enumerated domains can be expressed by a
tertiary relation between product class, property and
value as it is shown in figure 1. This data model
introduces elementary attributes for each entity type.
For instance, a property consists of its identification,
name, textual definition, data type, and unit of
measurement (UOM, e.g., meters, kilogram, and
volt). Similar models are used in (Bullig,
Schnadhorst & Wilkes, 2003), (Leukel, Schmitz &
Dorloff, 2002), and (Ondracek & Sander, 2003).
Property
PID
Name
Definition
UOM
Data Type
M
has_domain
VID
Name
Value
NM
CID
Name
…
…
1
N
has_subclass
Keyword
N
1
is_synonym_for
Product
Class
KID
Name
NM
described_by
Figure 1: Basic conceptual model.
The major drawback of this model can be
illustrated by a simple example: In industry segment
A, which is represented by a number of classes, the
property “length” is measured in inches, while
industry segment B – being represented by other
product classes of the same sub-tree – measures the
very same property in meters (e.g., hand tools vs.
pipes for gas transportation). Consequently, two
properties must be defined; their specification is
nearly equal and differs only in UOM. This is
especially for horizontal PCS, which cover a broad
range of industry segments, a common problem. It
can be solved by increasing the reusability of
properties, though this may require modifications of
the model.
4 MEASURES
In this Section, we define a comprehensive set of
measures that can be taken by standards makers. We
describe the rationale and, if necessary, point out to
extensions of the basic model. In addition, we assess
the impact of each measure on the given problem.
There are “two sides to every story” – standards
must be developed and maintained, and standards
should be applied; otherwise they are no standards.
The effect on standards makers concerns changes in
initial efforts (i.e., setting up the PCS) and
maintenance efforts (i.e., processing change
requests, releasing new versions). Similarly, the
effect on standards adopters refers to the initial
classification process and following re-classification
processes. However, an often overlooked aspect of
PCS adoption is the GUI presentation and actual
usage in applications such as e-procurement,
marketplaces, and product data management
systems. These issues will be considered, too.
4.1 Maximum Number of Properties
Rationale: The number of properties per class is
limited to a fix number (e.g., 15); this limitation
applies to all property lists, thus to the entire PCS. It
prevents property growth locally, especially with
regard to product segments in which product
descriptions can be very detailed. This measure does
not modify the basic model, but adds a constraint on
the cardinality of the class-property relation.
Standards makers:
The implementation of this
measure requires making a single decision on the
maximum number. Eventually, the size of the
property library is limited as well (only by the
number of classes and the reuse of existing
properties). There is even a significant change in
maintenance efforts, since adding a new property to
a property list is not possible if the maximum
number has already been reached.
Standards adopters:
Both the classification and
re-classification efforts are reduced and can be
forecasted. The GUI representation is improved due
to lower space requirements and may fit on a single
screen in all cases (e.g., imagine the list of 294
properties compared to the reduced list of the 15
most important properties).
4.2 Optional Properties
Rationale: Properties are distinguished whether their
usage is optional or mandatory. This distinction aims
at reducing the number of essential properties that
have to be used for product description, while it does
not remove properties from property lists. The
number of optional and mandatory properties
ICEIS 2006 - SOFTWARE AGENTS AND INTERNET COMPUTING
6
depends on the product class; for instance, all
properties may be optional, or mandatory (min/max-
approach). This measure adds a new attribute
“mandatory” to the class-property relation.
Standards makers: For each property of each
property list, the question of mandatory or optional
has to be answered. This initial effort can be reduced
by setting all properties to optional followed by
searching for the most important, thus mandatory
properties. Because of product innovation, optional
properties may be shifted to mandatory and vice-
versa (maintenance effort).
Standards adopters:
Depending on the share of
optional properties, the classification effort is
reduced. This measure allows diverse classification
strategies, i.e. support only mandatory properties.
Adopters decide on supporting optional properties,
especially if these are required by their customers.
Moreover, there are two consequences on GUI
representation: (i) optional properties can be marked,
thus separated from mandatory properties, and (ii)
parametric, property-based search for products has
to be restricted to mandatory properties.
4.3 Naming Conventions
Rationale: The name of a property must adhere to
specific naming conventions in order to prevent
redundant properties (e.g., “diameter, max.” vs.
“maximum diameter” vs. “max. diam.”). This
measure addresses the problem of finding the right
property in the property library, in addition to
keywords. Types of conventions: prefix vs. postfix
qualifiers, singular vs. plural, use of abbreviations,
separation of UOM from property name (e.g.,
“diameter” instead of “diameter in mm”). This
measure does not modify the basic model.
Standards makers:
Initially, naming conventions
must be developed, and applied to all property
names. Additional property names to the harmonized
name can be stored in the keyword list. Applying
these conventions can reveal redundant properties
that should be removed from the property library.
Standards adopters:
The classification process is
not directly affected; searching for the right property
is slightly improved. Since the naming can be used
to build logical groups of properties, the GUI
representation of large property lists is improved.
For instance, postfixes to property names (e.g.,
“length, max.”, “length, min.”) express a
specialization of closely related properties.
4.4 Property Groups
Rationale: Each property belongs to a predefined
group (e.g., design, dimensions, shape, and business
properties). This categorization eases the handling of
huge property lists, since the flat list is transformed
into a hierarchical structure. The basic model has to
be modified: define a list of groups, and add a N:1
relation between property group to property.
Standards makers:
Implementing this measure
requires defining non-overlapping groups and
assigning each property of the property library to
one group. The maintenance effort is slightly
effected (assign each new property to one group).
The subdivision of the property library helps
overlooking all properties, though it does not affect
the total number of properties.
Standards adopters:
Similarly to naming
conventions, this measure does not influence the
classification process. In the same way, it improves
the GUI representation by explicitly defined groups
of similar properties.
4.5 Views on Property Lists
Rationale: Instead of defining a single
comprehensive property list, define overlapping
views on property lists for each stage of the product
life-cycle. The rationale is that the relevance of a
property depends mainly on the product life-cycle
and the respective business function of product data.
For instance, the requirements of spend analysis
differ from those of engineering. In consequence,
each view-specific property list can be reduced to
purely relevant properties. Eventually, the class-
property relation must be modified to reflect the
view.
Standards makers:
Views on property lists result
in multiple, overlapping property lists for the same
product class; hence the initial and maintenance
efforts are considerably higher. While the number of
properties in those lists is reduced because of
including only view-relevant properties, the total
number of properties in the property library remains
unchanged.
Standards adopters:
Due to strictly view-relevant
properties, the efforts for classification can be
reduced in those cases where not all product
lifecycle phases are relevant. Often, standard PCS
are only used for procurement or sales and not for
intra-organizational purposes; therefore, this benefit
is quite relevant. In addition, the GUI representation
no longer contains non-relevant properties.
MAINTAINING PROPERTY LIBRARIES IN PRODUCT CLASSIFICATION SCHEMES
7
4.6 Property Templates
Rationale: Instead of defining properties completely,
the property library contains templates only. These
generic properties can be used for multiple specific
purposes by concretizing the template. This concept
is very similar to separating property definition and
property application as described in (Ondracek &
Sander, 2003). For instance, the template includes
name and definition, while the concretization adds
data type and UOM. Regarding the basic model, the
class-property relation is extended by further
attributes that were formerly part of the property
entity type.
Standards makers:
The first step for
implementing this measure is deciding which
attributes still belong to the generic property and
which attributes belong to the class-specific
property, thus to the class-property relation.
Eventually, a rather small number of generic
properties needs to be defined from which more
specific properties can be instantiated. Concerning
maintenance, adding a new property can often be
based on a similar, already existing property (i.e.,
concretizing the generic property instead of defining
the new property completely).
Standards adopters:
This measure does not
influence the classification process nor does it
improve the GUI representation. The reason is that
property templates concern only the organization of
the property library.
4.7 Property Inheritance
Rationale: So far, all measures were directed at
properties only. Considering that properties are
assigned to product classes forming a class
hierarchy, property inheritance says that properties
are inherited to all lower classes. Moreover, an
inherited property can be modified (concretized) on
lower levels. This measure does affect the basic
model as follows: the class-property relation as well
as the class-property-value relation is available for
all classes, not only for leaves of the class tree.
Standards makers:
Setting up a PCS based on
property inheritance calls for thoroughly defined
properties that can be assigned to nodes of the class
tree; otherwise the benefits of inheritance will not be
realized. Moreover, the class hierarchy itself has to
be suitable for assigning properties that are common
for complete sub-trees. Maintaining such a PCS
requires fewer efforts, since the property library
contains lesser properties and sub-trees truly
represent similar product classes characterized by a
set of common properties.
Standards adopters:
Similarly to property
templates, this measure concerns the property library
only.
5 DISCUSSION
In this Section, we discuss the proposed measures by
summarizing the expected effects, modifying the
basic model, and checking selected standard PCS to
which degree they take care of property growth.
5.1 Summarization of Effects
Next, we compile the previously assessed effects of
each measure on the problem of property growth
(see table 3). For both standards makers and
adopters, we state expected changes regarding initial
and maintenance efforts (“-“ for decrease; “+” for
increase, “o” for no change). Effects on the total
number of properties and GUI representation are
further criteria of our assessment (“reduced” and
“improved” respectively).
Comparing the effects on standards makers and
adopters, we have to state that 6 out of 7 measures
increase the initial effort for standards makers, while
classification efforts are reduced or remain
unchanged; a reduction of the number of properties
can be expected for four measures, while the GUI
representation is improved by five measures.
5.2 Modification of the Basic Model
Since the proposed measures concern the definition
Table 3: Effects on Standards Makers and Standards Adopters.
Measure Initial Effort Maintenance
Effort
Number of
Pro
p
erties
Initial Effort Maintenance
Effort
GUI
Maximum Numbe
r
- - reduced - - improved
O
p
tional Pro
p
erties
+oo - oimproved
Namin
g
Conventions
+ o reduced o o improved
Pro
p
ert
y
Grou
p
s
+ooooimproved
V
iews on Pro
p
ert
y
Lists
++o - -improved
Pro
p
ert
y
Tem
p
lates
+ - reduced o o o
Pro
p
ert
y
Inheritance
+ - reduced o o o
Standards Makers Standards Adopters
ICEIS 2006 - SOFTWARE AGENTS AND INTERNET COMPUTING
8
of single properties being elements of a property
library, the basic model for PCS needs to be
modified. Therefore, we collect the modifications
described before and alter the basic model as shown
in Figure 2. The modifications include (i) adding one
new attribute (mandatory), (ii) adding two new
entity types (property group, view), (iii) redefining
one relationship (described_by), and (iv) moving
attributes from the property entity type to the
described_by relationship (here: data type, uom).
Property
PID
Name
Definition
N
M
described_by
Mandatory
Data Type
UOM
View
VID
Name
M
M
has_domain
VID
Name
Value
NM
CID
Name
…
…
1
N
has_subclass
Property
Group
N
1
belongs_to
Keyword
N
1
is_synonym_for
Product
Class
KID
Name
Figure 2: Modified conceptual model.
5.3 Evaluation of Selected Standards
By analyzing existing standard PCS whether they
implement at least some of the proposed measures,
we reconstruct the current state of standard PCS
concerning the problem. We select five PCS for this
purpose: the two most important horizontal
standards (eCl@ss and EGAS; the latter adds
property lists to UNSPSC), and three vertical
standards (RNTD covers electronic and IT
components; ETIM and proficl@ss are vertical
European initiatives). For basic information on these
and many other schemes see (CEN/ISSS, 2004).
Table 4 shows the results of our evaluation.
We have to state that the coverage is quite little.
eCl@ss, which is often regarded as a thoroughly
developed PCS, implements one single measure
only: property groups by referring to a
categorization described in the IEC 61360 standard.
While no standard implements all measures, both
horizontal standards fall behind the vertical
standards. Naming conventions are provided and
followed by proficl@ss and RNTD, while the other
schemes do not describe conventions. However, due
to some of the actual property names it is probable
that implicit conventions exist (‘No’ in brackets
represents this). The support of the three most
sophisticated measures – views, templates, and
inheritance – is even lower with inheritance being
not implemented at all.
6 CONCLUSIONS
The main contribution of our paper lies in proposing
new measures for coping with the problem of
extensive and steadily growing property libraries
being prime examples of real-world business
vocabularies. The assessment of the impact on
standards makers and standards adopters revealed
that the effects concern not only the total number of
properties, but also initial efforts, maintenance
efforts, and GUI representation issues. Therefore,
decisions on implementing these measures should
bear in mind all these criteria. The results of the
assessment as well as the modified model may serve
standards makers in their decision process on
reorganizing property libraries.
The quantitative analysis (see Section 3.1) of the
property library in eCl@ss has drawn attention to the
problem of defining, maintaining and actually using
huge sets of product properties. While this quite
elaborated standard PCS claims to be unique in its
property-centered approach, the conceptual model of
its property library is rather simple. We conclude
that eCl@ss still focuses on semantic richness (i.e.
extending the coverage of industry segments) rather
than formal precision and efficient maintenance.
This example, nevertheless, underlines the need for
re-thinking the current organization of property
libraries, since measures tackling the described
problems are at hand, their impact can be predicted,
and some of the proposed measures have already
been tested in vertical standards.
Considering recent developments in finding
Table 4: Evaluation of Standard Product Classification Schemes.
Measure eCl
@
ss EGAS ETIM
p
roficl
@
ss RNTD
Maximum Number
No No No No No
O
p
tional Pro
p
erties
No No No Yes No
Namin
g
Conventions
(No) (No) (No) Yes Yes
Propert
y
Groups
Yes Yes No No No
V
iews on Propert
y
Lists
No No No No Yes
Propert
y
Templates
No No Yes Yes No
Propert
y
Inheritance
No No No No No
MAINTAINING PROPERTY LIBRARIES IN PRODUCT CLASSIFICATION SCHEMES
9
consensus about the basic components, underlying
conceptual models as well as maintenance policies
of PCS, standards bodies and industry consortia
have joined efforts in harmonizing their proprietary
approaches in several initiatives and on different
levels of obligation. For instance, the CEN project
on product classification (CEN/ISSS, 2005) states
that a “good” PCS necessarily incorporates
properties and property lists. Standard PCS that are
purely based on classes are expected to add
properties in order to broaden the range of
application and to provide extended semantics.
Concerning the current state of standard PCS,
these transformation processes will be quite
demanding. From this point of view, we plan future
work on validating the measures by quantitative
analysis of standard PCS, thus reengineering their
property libraries based on automated conversion,
and developing transformation strategies for
standards makers. Another field of interest derives
from the role and suitability of reference models for
PCS. These models will become more important
since many standard PCS aim at converting their
proprietary data models to the ISO 13584 standard
(ISO, 2001). This standard requires some
modifications to the property library, though it does
not address the maintenance problem explicitly as
described in this paper, since its main purpose is to
achieve semantic interoperability between different
property libraries. Therefore, we see the need for
extending the scope of this reference model to
content management issues that greatly determine
the costs and efforts of defining and implementing
respective standard PCS.
REFERENCES
Beneventano, D. et al., 2004. A web service based
framework for the semantic mapping amongst product
classification schemas. In Journal of Electronic
Commerce Research, Vol. 5, pp. 114-127.
Bullig, A., Schnadhorst, T., Wilkes, W., 2003. Mapping of
product dictionaries and corresponding catalog data. In
CE 2003, 10th ISPE International Conference on
Concurrent Engineering. Balkema, pp. 225-234.
CEN/ISSS, 2004. CEN Workshop Agreement 15045:2004
– Multilingual catalogue strategies for eCommerce
and eBusiness. Brussels, Belgium.
CEN/ISSS, 2005. CEN Workshop Agreement 15925:2005
– Description of References and Data Models for
Classification. Brussels, Belgium.
Corcho, O., Gómez-Pérez, A., 2001. Solving Integration
Problems of E-commerce Standards and Initiatives
through Ontological Mappings. In Workshop on E-
Business and Intelligent Web at the 17th International
Joint Conference on Artificial Intelligence.
eCl@ss e.V., 2004. eCl@ss – Standardized Material and
Service Classification, Version 5.1. Retrieved January
23, 2006 from http://www.eclass-online.com.
Fairchild, A. M., Vuyst, B. de, 2002. Coding Standards
Benefiting Product and Service Information in
Ecommerce. In HICSS-35, 35th Annual Hawaii
International Conference on System Sciences. IEEE
Computer Society, pp. 258b.
Fensel, D. et al., 2001. Product Data Integration in B2B E-
Commerce. In IEEE Intelligent Systems, Vol. 16, pp.
54-59.
Hepp, M., 2004. Measuring the Quality of Descriptive
Languages for Products and Services. In MKWI 2004,
Multikonferenz Wirtschaftsinformatik 2004. Cuvillier,
pp. 157-168.
Hepp, M., Leukel, J., Schmitz, 2005. A Quantitative
Analysis of eCl@ss, UNSPSC, eOTD, and RNTD:
Content, Coverage, and Maintenance. In ICEBE 2005,
IEEE International Conference on e-Business
Engineering. IEEE Computer Society, pp. 572-581.
ISO, 2001. ISO 13584-1:2001 Industrial automation
systems integration – Parts library – Part 1: Overview
and fundamental principles. Geneva, Switzerland.
Kim, D. et al., 2004. A semantic classification model for
e-catalogs. In CEC 2004, IEEE International
Conference on E-Commerce Technology. IEEE
Computer Society, pp. 85-92.
Leukel, J., 2004. Standardization of Product Ontologies in
B2B Relationships – On the Role of ISO 13584. In
AMCIS 2004, 10th Americas Conference on
Information Systems. AIS, pp. 4084-4091.
Leukel, J., Schmitz, V., Dorloff, F.-D., 2002. A Modeling
Approach for Product Classification Systems. In
DEXA 2002, 13th International Workshop on
Database and Expert Systems Applications. IEEE
Computer Society, pp. 868-874.
Ng, W., Yan, G., Lim, E.-P., 2000. Heterogeneous product
description in electronic commerce. In ACM SIGeCom
Exchanges, Vol. 1, pp. 7-13.
Ondracek, N., Sander, S., 2003. Concepts and benefits of
the german ISO 13584-compliant online dictionary
www.DINsml.net. In CE 2003, 10th ISPE
International Conference on Concurrent Engineering.
Balkema, pp. 255-262.
Schulten, E., 2001. The E-Commerce Product
Classification Challenge. In IEEE Intelligent Systems,
Vol. 16, pp. 86-89.
ICEIS 2006 - SOFTWARE AGENTS AND INTERNET COMPUTING
10
