Ontologies have been used as vehicles to assign semantics to data in several domains, since they represent a conceptualization about the domain of interest and establish a common vocabulary to be shared. For properly addressing the domain of interest and serving as a means to attribute meaning to data, ontology should provide a correct understanding of the domain conceptualization. This paper presents some results of an ontological analysis of the Good Relations Ontology (GRO). For the ontological analysis, we used Goal Modeling to delimitate the ontology scope and the Unified Foundational Ontology to analyze its conceptual model. Our analysis discovered several limitations that prevent GRO from properly addressing the domain of interest. Thus, we reengineered GRO to correct such limitations, and implemented it using OWL.
Trading over the Internet is a marketing model that has increased in the last years and tends to be the major trading method in the near future. For succeeding in this market, organizations have to make their products and services visible to potential customers. Having product and service offerings information captured by search engines, recommender systems and other applications can increase the visibility of product and service offerings and promote competitiveness. The use of ontologies to assign semantic to data is useful in this context.
The Good Relations Ontology (GRO)1 is a lightweight ontology addressing the e-commerce domain, particularly product and service offering. It is basically a domain structural model of generic concepts that can be used as a schema to describe ecommerce related data, such as products, prices, and product properties. It provides a common vocabulary that can be embedded into existing static and dynamic web pages and processed by computers. It is used by several companies (e.g., Google, Kmart, BestBy) and allows for searches involving offerings and suppliers.
The quality of the ontology model is crucial for gathering consistent information
from data linked to the ontology
Considering that GRO has been used as the e-commerce data model of the
Schema.org2 initiative and has a lot of data linked to it, we decided to evaluate GRO
taking the aspects previously mentioned into account. Aiming to understand the GRO
domain of interest and, thus, the GRO scope, we applied Goal Modeling. Goal
Modeling defines procedures to identify and express domain requirements in software
engineering and it has been proposed as a method to establish ontology scope by means
of the competence questions that the ontology should be able to answer
This paper addresses the GRO ontological analysis and presents some of the produced results. It is organized as follows: Section 2 provides the background for the paper, talking about Goal Modeling, UFO and OntoUML; Section 3 addresses the use of Goal Modeling to understand the GRO scope; Section 4 describes GRO ontological analysis and redesign; and Section 5 presents our final considerations.
Goal modeling has been successfully applied in requirements engineering
When applying goal modeling, it is necessary to define the framework or
terminology used to express the goal models. The i* framework
In the ontology engineering context, goal modeling can help understand the
domain of interest and the involved actors. Moreover, from goal models it is possible to
derive competency questions to delimit the ontology scope.
The Unified Foundational Ontology (UFO)
Figure 1 depicts the endurant universals hierarchy in UFO. An Endurant
are kinds of universals whose instances are Substantial Individuals and Moments, respectively. Substantials are existentially independent entities (e.g., John). Moments, in turn, are entities that can only exist by inhering in other individuals (e.g., color). A Moment can be a Mode, which inheres in a single individual (e.g., color) or a Relator, which depends on a plurality of entities (e.g., employment) and, for this reason, provides the material connection between them.
Substantial Universals are distinct regarding principle of identity, rigidity and dependence. Concerning the principle of identity, the distinction is between Sortal and Mixin Universal. The instances of a Sortal have a unique principle of identity while Mixin Universals are abstract entity types that have instances with different principles of identity. For instance, John is an instance of Person (Sortal Universal) that obey a unique principle defined in function of Person’s essential attributes. In this case, Person provides the principle of identity to its instances. John can be also an instance of an abstract entity type Client (Mixin Universal) that can have instances that are Person or Organization and, thus, have different principle of identity. Thus, Mixin Universals do not provide any principle of identity for its instances, instead, they just aggregate properties of distinct principle of identity providers. Both Sortals and Mixin Universals are classified in terms of rigidity. Rigidity is a modal property that defines the behavior of the instances of a given entity type. An entity type is said rigid iff its instances never cease to be its instances in all possible worlds (e.g., Person, Chair). Otherwise (if some instances of an entity type can cease to be instance of this type in a possible world), this type is said non-rigid. An entity type is anti-rigid if it is non-rigid and if all its instances necessarily can cease to be its instances in a possible world (e.g., Student, Employee). Substance Sortals (Kind and Collective) are rigid entity types that provide principle of identity for theirs instances. A Kind represents a functional complex which means that it has distinct parts (e.g., Human Being) while a Collective is an entity type that is composed of equal parts (e.g., Committee). All other Sortals Universals (i.e., Role, Phase and Subkind) are entity types that can only occur as specializations of a Substance Sortal, since all instances must have exactly one principle of identity provided by a unique Sortal. Thus, a Subkind is a Rigid Sortal Universal that is a specialization (directly or indirectly) of a Kind or a Collective. Nonetheless, Role and Phase are anti-rigid types differentiated by the dependence aspect. A Role Universal is externally dependent, which means that its instances necessarily depend on another instance (e.g., System Analyst). A Phase Universal is defined by intrinsic properties (e.g., Child). The distinction between Mixing Universal types only regards rigidity. Then, a Category is rigid and RoleMixin is anti-rigid.
The UFO-A categorization was used as a basis to OntoUML
In order to evaluate GRO, it is important to understand its domain of interest and, thus, its scope. Therefore, we applied Goal Modeling to represent the main aspects and goals related to the domain. It is important to point out that, although GRO is said to be an ontology for the e-commerce domain, in fact, it addresses aspects related only to offering. More specifically, only to product offering (further discussions about that are made later). Thus, we applied Goal Modeling to represent knowledge about product web-offering and establish the GRO scope.
We started by analyzing a number of offerings available on the web and some literature about the topic aiming to gather knowledge about relevant information regarding web-offerings. Then, we identified the main actor (Offering Specifier) and stakeholders involved in the domain (Shipping Company, Payment Company, Delivery Company and Business Warrantier) and built the Strategic Dependency model (not shown here due to space limitation) to identify the relations between them. After that, we identified the main actor goals to understand his needs. The main goal of an Offering Specifier is have the offering specified. In order for him to accomplish this goal, it is necessary to accomplish the following ones: have the price specified, have the offering description specified, have the delivery or shipping methods specified, have the payment methods specified. Once the actor’s goals were identified, for each goal, we derived tasks to accomplish it and identified the necessary resources to perform them. Figure 2 shows a fragment of the Strategic Rationale model produced.
For achieving the have the offering description specified goal, the Offering Specifier needs to perform the following tasks: inform the offering type (e.g., sell, rent) and inform offering (which includes offering item, offering price specification, offering characteristics (e.g., textual description, pictures), offering item categories and subcategories, payment methods, shipping or delivery methods, provider, delivery time, location available (e.g., Worldwide, Europe), provider location and delivery charge). For achieving the have the payment methods specified goal, the Offering Specifier needs to inform payment methods and for achieving the have the delivery or shipping methods specified, the Offering Specifier needs to inform shipping or delivery methods.
After developing the goal models, we followed the method proposed by
Since these competency questions describe the web-offering domain, they
should be answered by any ontology addressing that domain. GRO is claimed to be an
e-commerce lightweight ontology and, since the web-offering domain is part of the
ecommerce domain, GRO should be able to answer those questions. However, GRO does
not answer several of them. Moreover, by analyzing the GRO competency questions
Additionally, the GRO competency questions are described in a poor manner.
Some of them mix several questions into only one sentence or provide a lot of options
in the same sentence. For example: Which retrievable Web Resources describe an offer
{to sell | to provide the service of | to repair | to maintain | to lease out | to dispose} {a
concrete individual | some unknown individuals} of a {given good | given service |
spare part for a given good | consumables and supplies for a given good} described by
a {type of good | specific make and model} that meet certain requirements on
{properties | intervals for properties} for which the offering party accepts a given
method of payment and provides a certain method of delivery to {consumers |
retailers} in a given {country | region}?
The competency questions derived from the goal models delimit the scope of a web-offering ontology. By comparing the delimited scope with GRO scope we noticed that several aspects are not addressed by GRO. After understanding the web-offering domain and analyzing the GRO scope, we analyzed the GRO conceptual model. The main results are discussed in the next section.
Our evaluation of the GRO conceptual model regards two different aspects. In the first, we analyzed the model representation and domain coverage. Then, we evaluated the GRO model, verifying if it can properly answer the competency questions identified from the goal models. Finally, we redesigned the GRO model using OntoUML. Due to space limitation, we do not reproduce the GRO model in this paper. It can be obtained in http://www.heppnetz.de/projects/goodrelations/.
GRO is implemented in OWL. Its model is represented by using UML and does not represent the associations cardinality. OWL models, in general, do not apply any mechanism to forbid invalid cardinalities and this may be the reason why the GRO UML conceptual model does not represent cardinalities. However, in order to be properly used for both specifying offerings and processing data linked to GRO, it would be desirable to know the cardinalities. They are important, for instance, to distinguish mandatory from optional associations. For example, an offering should always specify the locations where it is available. Thus, the minimum cardinality from offering to location must be 1 (i.e., a mandatory relation). Contrariwise, there can be products with no brand (i.e., optional relation).
There are problems in some attributes defined in GRO. For instance, what GRO defines as ProductOrService, and we have decided to call Offering Item, has the Color attribute. However, a service does not have a color. Moreover, neither a product always has a color (e.g., virtual products like music files, books, etc.). Additionally, a product can have many colors, and in GRO this attribute accepts only one value. We do not agree to define Color as an attribute. We consider color as an instance of Qualitative Property. This way, it would only be used when an offering item has color and it would be possible to associate an offering item with many colors.
There are also problems regarding the Category attribute, which is defined as a literal. We can find on the web many examples of offerings specifying items that belong to many categories and subcategories. Thus, a literal is not a proper way to specify offering items since it turns complicated, for example, to find instances of the same category by comparing literals. For this reason, we argue that Category should be a class instead of an attribute and its instantiation should be done by the Product Types Ontology3, which defines a huge and precise dictionary of product types and brand names for GRO. In this case and in others like this, the integration with other ontologies could increase GRO quality. For instance, GRO has a class to define the locations where an offering is available. This class has attributes like Name and Category as literals. Once more, it is not a good way to specify locations and location’s category as literals, since it might be useful to apply this information in a structured way. We argue that in these cases, GRO should link to other ontologies (e.g., the Location Ontology4) that address the concepts more suitably.
In order to verify domain coverage, we checked if GRO is able to properly answer the competency questions presented in the Section 3. We classified the CQs in four groups. The first one refers to CQs that GRO can answer. The second comprises the CQs that GRO does not address. The third includes CQs that could be better answered by integrating GRO with other ontologies. The last refers to CQs that cannot be answered due to problems in the GRO model. Table 1 summarizes the CQs of each group. 4 http://dbpedia.org/ontology/Location GRO can answer GRO does not address Other ontologies are necessary GRO cannot answer due modeling problems 5, 7, 8
As one can see, although GRO can answer many CQs, it disregards several important aspects related to offering specification, like the offering item components. Moreover, a modeling problem prevents GRO from answer CQs 5, 7 and 8. This occurs because of the class that GRO calls ProductOrService and that we understand as Offering Item. In the GRO Model this class generalizes Individual (i.e., Individual Product) and Product Model. The problem with this representation is that since Product Model is a subtype of Offering Item, a Product Model would be always a ProductOrService (i.e., an offering item). This is not the case when the ProductOrService is an Individual Product (e.g., a particular cellphone) and the Product Model (e.g., iPhone 7) is just information about the Product being offered and, thus, it is not a ProductOrService). This problem leads to inconsistent instantiations that impacts in the CQs mentioned.
A conceptual model is intended to represent a domain with the purpose of
communication
Since OntoUML is based on UFO categories, it has a number of constructs to better specify domain conceptualizations. In other words, since it applies a number of philosophical and cognitive theories extending the UML syntax, it can offer tools to enrich the semantic of the GRO model and solve the aforementioned problems. Next, we present fragments of the redesigned GRO conceptual model, discussing the advantages brought by the use of OntoUML. In the figures, stereotypes indicate the OntoUML categorization.
OntoUML has a number of constraints that must be respected to use each
categorization (e.g., a Role must have a Kind as supertype). To build a consistent
OntoUML model that does not violate these constraints, the OntoUML patterns
presented in
By applying the Role Pattern
Figure 4 answers CQ9, a competence question not addressed by GRO. To express the partOf relation between the Offering Item and its components, we applied the OntoUML componentOf relation. This relation applies between a functional complex and its parts. Therefore, the functional complex Offering Item contains a set of parts Offering Item Components. It is important to point out that we consider as Offering Item a set of textual descriptions as it is represented on web-offerings (e.g., the description of a car engine), and not the object being described (e.g., a car engine). The main issue here is that an Offering Item may have components that can have recursively other components. This can be solved in OntoUML by applying two subkinds of Offering Item Component, such that, one represents an atomic structure (i.e., a component that has no components) and the other a complex one (i.e., a component that has other components). A complex component can be composed by atomic and complex components and, thus, it has the componentOf relation associated to the generic type.
After redesigning GRO conceptual model, we rebuilt the GRO OWL model. For
that, we applied an automatic transformation from OntoUML to OWL presented in
Ontological analysis is an important tool to promote improvements in conceptual
models and ontologies. This paper presented some of the results of an ontological
analysis and reengineer of the Good Relations Ontology, carried out using Goal
Modeling and OntoUML. We applied Goal Modeling to understand the GRO scope and
domain of interest. From goal models we derived competency questions that should be
answered by an ontology addressing the web-offering domain. Aiming to improve the
GRO conceptual model, we redesigned it using OntoUML and, then, we developed an
OWL ontology. By ontologically analyzing GRO model, we included a foundational
layer beneath the concepts and relations, providing the needed ground and consistency
to improve GRO towards an e-commerce high quality ontology
Several works have addressed ontology evaluation, attempting to give
ontological foundations to conceptual models. Regarding GRO, a few studies were
conducted in order to evaluate it.
The main contributions of the work addressed in this paper are: (i) goal analysis and definition of competence questions for the web-offering domain, aiming to improve GRO scope clarity; (ii) evaluation and redesign of the GRO conceptual model using OntoUML; and (iii) a new OWL ontology model aligned to GRO. As future works, based on the results of this work, we intend to develop an e-commerce ontology and use it in interoperability initiatives.
This research is funded by the Brazilian Research Funding Agency CNPq (Processes 485368/2013-7 and 461777/2014-2) and FAPES (Process 69382549/2014). 5 https://www.w3.org/TR/owl-time/#time:DateTimeDescription