The GoodRelations Ontology is experiencing the first stages of mainstream adoption, with its appeal to a range of enterprises as the eCommerce ontology of choice to promote its product catalogue. As adoption increases, so too does the need to review and analyze current implementation of the ontology to better inform future usage and uptake. To comprehensively understand the implementation approaches, usage patterns, instance data and model coverage, data was collected from 105 different web based sources that have published their business and product-related information using the GoodRelations Ontology. This paper analyses the ontology usage in terms of data instantiation, and conceptual coverage using SPARQL queries to evaluate quality, usefulness and inference provisioning. Experimental results highlight that early publishers of structured eCommerce data benefit more due to structured data being more readily search engine indexable, but the lack of available product ontologies and product master datasheets is impeding the creation of a semantically interlinked eCommerce Web.
The Web of data and open ontologies (e.g. FOAF, SIOC, SKOS)
promotes the establishment of a shared understanding between
data providers and consumers in a common format that allows the
automated processing of information by software agents. Where
accepted by the community, an ontology offers the opportunity for
enhanced dissemination and commercial use of information. The
GoodRelations Ontology (GRO) [
PingTheSemanticWeb.com1 has ranked GRO second to FOAF as the most widely used ontology. Available since 2008, GROs’ schema is mature but uptake reflects that of early adoption. A review and analysis of the current community implementations of the GRO within its eCommerce environment is timely as it will provide insight into its applicability, conceptual coverage and actual usage within its application domain. This paper reports on the current implementation status of the GRO after investigating 105 publically available data sets. In this first large scale investigation of its kind into the GRO, data providers are categorized, dataset characteristics discussed and the usefulness of currently available data sets analysed through different use cases. Implicit data available through axiomatic triples is also considered.
The remainder of the paper is organized as follows. Section 2 introduces the motivation, and background is discussed in Section 3. In Section 4, we discuss the dataset collection and its characteristics. Section 5 describes the dataset investigation and use cases, along with results, observations and impact of reasoning. Related work is presented in Section 6 and Section 7 concludes the paper.
The semantic web provides a level of semantically annotated
structured data that enhances the level of user experience by more
accurately sourcing and identifying information of interest.
Enabled primarily through ontological alignment, semantic
annotation is a major factor contributing to the increasing interest
in ontology usage by the wider community and one which had
also attracted the attention of early business adopters. Over the
last two years, the GRO has witnessed this sectoral appeal with
mainstream adoption by eRetailers such as BestBuy.com,
Overstock.com and Oreilly.com. Announcements from search
engine providers Google2 and from Yahoo3 to index the GRO
will, for its corporate users, extend consumer reach to a larger
audience with their increased appearance in search results. A
measure of the popularity of any ontology is its community
acceptance, which reflects ‘some’ level of use but not the extent
of adherence to the schema or the extent of instantiation. A more
accurate examination of popularity should therefore consider the
1 Last accessed on Dec 16, 2010
2http://googlewebmastercentral.blogspot.com/2010/11/rich-snippets-forshopping-sites.html
3 http://developer.yahoo.com/searchmonkey/smguide/gr.html
overall ontology population. To date, however, the literature does
not present evidence of systematic analysis of GRO usage that
could provide this insight into its adoption and usage status in the
emerging eCommerce Web of data.
[
In the following section, we describe our high level categorization of data providers, and present a brief overview of the GR conceptual schema and use of the GRO in search indexes.
Looking at the structured eCommerce data landscape, we can categorize users into three groups based on their publishing approach, usage pattern and data volume.
This group includes large online e-retailers and retailers who are traditionally premises-based and have only recently entered the eRetailing business. Such data sources provide more detailed (rich) product description which is useful for entity consolidation and interlinking with other datasets. Such companies include BestBuy.com, Overstock.com, Oreilly.com, and Suitcase.com.
A large number of the data sources included in our dataset comprises small to medium web shops, offering their products and services mainly through web channels. Most of these web shops use web content management packages4 such as Maganto, osCommerce and Joomla to add RDFa data in html pages. This approach works well since no special infrastructure arrangement is required in most cases.
To leverage the benefits offered by semantic eCommerce data, businesses are offering data services that build on consolidated semantic repositories. Moreover, the providers use APIs to access and transform proprietary data into RDF before making them available through their repositories. For example, Linked Open 4 Complete list of their references are available at http://www.ebusinessunibw.org/wiki/GoodRelations#Shop_Software Commerce (LOC)5 contains Amazon.com Amazon.com has not yet published RDF/RDFa. data although
The latest version6 of the GRO comprises 27 concepts (classes),
49 object properties, 43 data properties and 43 named individuals.
To cater for backward compatibility, the ontology model was
recently updated with the addition of new object and data
properties based on implementation feedback. Note, gr is the
GRO prefix used in general practice and throughout this paper.
For the full specification of the GRO, the reader is referred to
http://purl.org/goodrelations/v1.
3.2.1 Axioms
The GRO comprised classes, properties, individuals and axioms.
Axioms allow information to be inferred from a knowledge base
through the use of a reasoning engine known as a reasoner. The
expressivity of the GRO is based on an OWL DLP fragment and
contains subclass and subproperty axioms to express the
subsumption behaviour in the model. Axiomatic triples in the
GRO are given in Table 1 to shed light on the possible inference
on eCommerce data which has been annotated using the GRO and
applicable rule sets. RDFS and OWL elements such as
rdfs:domain and rdfs:range, which are available in the ontology,
are omitted from the table as they were not included in the
reasoning experiment.
Elements such as rdfs:subClassOf, rdfs:subPropertyOf,
owl:inverseOf, owl:TransitiveProperty and
owl:SymmetricProperty were considered as they are associated
with new knowledge. They can be used both in forward-chaining,
to materialize the implied statements thereby making them
explicit, and in backward-chaining performing query rewrites to
expand query scope and include inferred knowledge.
owl.DisjointClasses differs because it is used primarily for data
quality and checking for inconsistencies. Constructs mentioned in
Table 1 are covered by almost all of the rule sets including RDFS,
pD* [
6 Latest version and model used in our experiment last updated on Nov 26, 2010.
The adoption of the GRO is driven by the level of enhanced
visibility that a company’s products and general profile can
receive as a result of its GRO marked-up data being included in
the search engine indexes of large providers such as Google and
Yahoo [
The eCommerce data sets constructed were collected primarily from those annotated with the GRO and represent the maximum number of accessible data sets. Throughout this paper, we use GoodRelations Dataset or GRDS to refer to the RDF graph collected from the various websites and stored in a triple store for querying and reasoning, and Data Source to refer to the websites’ unique domain name server (DNS) included in GRDS, which contains eCommerce data in RDF (any serialization format) or an RDFa format based on the GRO model.
To analyse the adoption, usage patterns and uptake of the GRO in general, and by the eCommerce community in particular, data sets were collected from multiple sources and consolidated to create the GoodRelations Dataset9 (GRDS). Potential data sources that used the GRO to describe the offerings or company (Business Entity) were the primary identification drivers. Different semantic search engines such as Sindice10 and Watson11 which index RDF documents were used to obtain a list of potential data sources. Traditional search engines such as Google were also used to retrieve RDF documents by using the filetype:rdf attribute of advanced search to access RDF documents over the web. Additionally, we also considered the list of data publishers maintained at the GoodRelations’s developer wiki site12. For our empirical investigation, data was collected from 105 different data sources complying with the stated criteria. The complete list of these data sources is provided in the Appendix.
During the collection process, we noticed that 90% of the websites (data sources) were using the RDFa13 standard to add structured information to existing HTML documents. The majority of sources had sitemap.xml files that allowed search engines to crawl the web pages and build indexes. However, the links (URLs) provided in the sitemap files were often linked to a list of products pages and not to the actual product pages themselves. Being interested in accessing the web pages that had 8http://www.nytimes.com/external/readwriteweb/2010/07/01/01readwriteweb-howbest-buy-is-using-the-semantic-web-23031.html 9 http://debii.curtin.edu.au/~jamshaid/GRDS-dump-v0.1.rar 10 http://www.sindice.com 11 http://watson.kmi.open.ac.uk/WatsonWUI 12 http://www.ebusiness-unibw.org/wiki/GoodRelations 13 http://www.w3.org/TR/rdfa-syntax/ embedded RDFa code, we relied upon crawlers to build the list of such URLs before manually verifying the constructed list. With a valid list of URLs REST-based web services, Any2314 and RDFa Distiller15 were used to parse RDFa snippets from online HTML documents and generate RDF graphs (in RDF/XML syntax). Graphs were then loaded into the OpenLinks Virtuoso16 triple store to create the GRDS experimental data set. From an RDF data management perspective, named graphs are used to group triples from one data source under a uniquely named graph URI, allowing the dataset to be queried vertically and horizontally. Linked Open Commerce17 (LOC) represents an emerging data space which collates eCommerce data from the Web and makes information available for retrieval and viewing through a SPARQL endpoint. Despite its presence, collection of data sets within this environment proved difficult owing to: i) the unavailability of several data sources in the LOC; and ii) the presence of several triples using non-authentic URIs, resulting in an inability to de-reference the URI, use it in a query or obtain provenance details. The LOC environment has approximately 34 data sources which publish and make their data available in RDF/RDFa format. Moreover, LOC contains a nominal number of data sources that are made available in RDF/RDFa but through the use of middleware APIs’ such as Amazon.com. Invalid URIs such as those starting with “localhost.localdomain….” were also found18 to be problematic, but overall represented minor data quality issue that was easily overcome.
The inclusion and use of these two datasets (i.e. LOC and GRDS) in our experiment provides the best opportunity for an optimal search space covering the maximum possible width (GRDS) and depth (LOC) of the structured eCommerce web-of-data. In essence, GRDS covers a greater number of data sources while LOC has greater coverage of data from data sources. Hence, both datasets complement one another with LOC providing an ability to cross check or find additional information which is useful for the analysis of results.
During GRDS data set composition, the different characteristics of the datasets such as use of different namespaces, GR vocabulary and annotation properties were considered. The results from each investigation are described below.
Table 2 lists all vocabularies and their prefixes found in GRDS. Apart from gr, the top three most used vocabularies are dc (Dublin Core), foaf and vCard. Some vocabularies, such as vCard and dc were found to be used with multiple prefixes due to the availability of a new version with new namespace URIs. A larger percentage of focused vocabularies were used by data sources to annotate the data relevant to their businesses; for example, frbr is used by O’Reilly to annotate bibliographic data. 14 http://any23.org 15 http://www.w3.org/2007/08/pyRdfa/ 16 Open Source Edition, http://sourceforge.net/projects/virtuoso/ 17 http://linkedopencommerce.com 18 Observations were made as of 16 OCT 2010
Here, an analysis of the GR vocabulary usage by different implementers was undertaken. The straightforward approach was used to calculate the number of instances each concept has in the dataset and calculate their properties used in implementation. While this approach helps to identify both the most and least populated terms, it does not provide sufficient understanding about usage patterns across different data sources. For example, if one particular class is used by a large implementer, e.g. BestBuy.com for their two hundred thousand plus products, then the count of instances of that class will be high. It is also equally possible that this particular implementation has used this concept in GRDS. Therefore, we consider the usage of ontology terms based on the percentage of data sources where it is in fact used, rather than on the total number of triples in the dataset. Our study also analysed the usage of GR schema from the perspectives of concept usage related to the instances and data sources found within the GRDS. 19 W3C has now RDF based vCard however 25.71% of data source are still using deprecated namespace 20 Yahoo search monkey project defined these namespaces to provide vocabulary to assist developers. 21 This is Google vocabulary published to be used for structured data with RDFa and microformat. 22 Facebook Open Graph protocol. Only used by www.lovejoys-ltd.co.uk 23 Vocabulary for expressing reviews and ratings. Only used by www.overstock.com 24 Vocabulary for Functional Requirements for Bibliographic Records (FRBR). Only used by www.oreilly.com 25 Vocabulary for latitude, longitude and altitude in the WGS84 geodetic reference datum. Only used by www.bestbuy.com This allowed a basic understanding about the nature of available data and the frequency of concept and/or property use by different data providers. However, statistical representation based on simple instance and data sources calculation does not provide insight into the relationships that exist between entities and the implementation of the ontological model in practice. To achieve this level of visibility, we investigated the level of GR model usage by examining the conceptual coverage of three main pivotal concepts (Business Entity, Offering, Product or Service) and description richness available by exploring (traversing) the relationships available with other concepts through GR properties (see Section 5). Figure 1 positions concepts on the diagram based on the percentage of their use across different data sources. Concepts are shown as the two groups of :Offering26 and :BusinessEntity, with the groups intended to assist with visualizing the specific content and use of a particular data fragment. Several concepts appear on the edge of the outermost circle, indicating that several data providers have not made available any fine grained information about their offerings, although they have provided basic data sets for eligible customer types or business functions for activities such as selling, leasing or renting. The lower half of Figure 1 details concepts linked directly or indirectly to the business entity (:BusinessEntity). Overall, 60% of the data sources provided information regarding their stores (office/branches) and 40% of the data sources have made further details available such as the opening and closing times of stores. One of the pivotal concepts i.e :ProductOrService is not shown in Figure 1 as no formal product ontology is currently being used by GRDS. It was found that the :ProductOrService subconcepts were used throughout the 26Throughout this paper, we assume that http://purl.org/goodrelations/v1# is the default namespace and use prefixes mentioned in Table 2 for other namespaces offering data to describe whether the product referred to in the offering is the actual instance or existentially quantified.
GRO recommends the use of annotation properties to provide additional information about resources. Almost all the entities in GRDS are annotated with rdfs:label and rdfs:comment properties. eCommerce deployment frequently used these in queries to retrieve resources of interest due to the properties being highly usable. For example, one of the instance of type :Offering has rdfs:label set to “13 pieces of product "Cash Bases Cost Plus Flip Lid 460, weiss" are on stock”. One possible solution is to use “Lid 460” in the FILTER clause of the SPARQL query to limit the result set to potential candidate offers. As indicated, en27 (IETF's BCP 47 code of English Language) is the natural language most commonly used for providing textual description of the resources. 27 http://www.w3.org/International/articles/language-tags/
One of the main purposes of making structured data available on the semantic web is to allow users to access accurate (exact) information [9]. The key to accessing exact information is the availability of a conceptual description based on the ontological model. The GRO contains concepts and descriptions that help in the publishing and consuming of eCommerce data on the web. We investigated the GRDS by considering common eCommerce use and observed the data response to these requirements. Following the GR conceptual model and focusing on pivotal concepts, we issued targeted queries against the dataset and analysed the results. In our investigation, we firstly analyzed the overall conceptual coverage of the model in order to understand the data landscape. Secondly, we performed a focused analysis so as to understand the richness of data in GRDS. As part of the focused analysis for each use case scenario, we firstly discussed the common understanding of concepts and the set of basic questions one can ask of the dataset. Secondly, queries were constructed from these questions to retrieve information and provide a better understanding of data. Finally, an analysis was conducted of each use case.
To understand the overall distribution of data and the conceptual coverage of the GR model in GRDS, different queries in different combinations were used. Figure 2 depicts the results in chart format. The y-axis represents the number of data sources, and xaxis shows the number of used queries (the queries are listed in Table 5). The shaded area reflects the information space available in GRDS. For example, point 6 of x-axis shows the number of data sources which provided data for the concepts listed in 6th row of Table 5. The query used for 6th row is available in listing 1. The highlighted area in the chart details the type of structured information currently available in eCommerce. Broadly speaking, we observe that, on average, every data publisher has provided business entity, offering and price details. However, almost no data source has provided any formal specification of the products being offered.
Listing 1: Query (representing the concept involved in point 6 of chart’s x-axis)
As previously mentioned, we use generic use case scenarios to illustrate the extensive use of semantic eCommerce data.
Finding a company is a very common and useful requirement in multiple situations, particularly when seeking a company in a specific vertical industry, a company offering a specific product, a company with a specific business role (buyer/seller), or even competitors. Intuitively, one could ask many questions to obtain the required information from the eCommerce information space. We have intentionally limited our search to the following questions as they are very basic and cover most user requirements. - Find a company with a specific name - Find a company in a particular location - Find a company in a particular line of business (or service) These questions also contain basic parameters that, if used in different combinations, can address more advanced requirements. To obtain a view of the structured information published by different data providers, we accessed the GRDS using SPARQL query shown in Listing 2.
Listing 2: Query (retrieving company description) Result and Observations With reference to Table 6, within GRDS, 93.34% of the data sources provided a business name using the :legalName property. This property is very helpful when searching for a company with a specific name using the SPARQL filter option. A few data sources28 were found which did not supply a value for the legal name property. Further investigation of these providers’ datasets found the presence of rdfs:label, vcard:fn properties but also with no attribute value. The unique identification of a Company (:BusinessEntity) on the Semantic Web using a string value is complicated as multiple companies often have the same name. Entity disambiguation [10] is required to distinguish identically named companies from each other. Despite the fact that the GRO has useful attributes that assist in identifying a company easily and accurately, we found only one data source29 in the GRDS that provided both the :ISICv4 code value (i.e. 4652) and company name. We did not find any value for the other predicates mentioned in the OPTIONAL clause of the SPARQL query above (see Listing 2). In the GRDS, the second-most widely used schema (after GRO) is vCard30 which provides the location and specific address of a company or shop. 99.5% of the data sources provided information 28 www.sachse-stollen.de, www.golfhq.com, ww.hagemann24.de, www.globalautoimports.com.br, www.xtremeimpulse.com www.cardgameshop.com, ww.discountofficehomefurniture.com 29 www.jarltech.com about the country and locality, and 85.3% also provided a street address with postcode (postal address).
Location of Store31, indicating where the service/product is provided, is annotated using the :LocationOfSalesOrServicesProvisioining concept . It has relationships with both :BusinessEntity (through :hasPOS) and :Offering (through :availableAtOrFrom). This allows information about the shop to be accessed by referring to the Business Entity or Offering. Shop location-related information is very helpful in many situations such as when visiting the shop, requesting online delivery or searching for a particular item in a particular location. In GRDS, 71.42% of data sources have provided shop information using :availableAtOrFrom and 44.76% have provided shop information using :hasPOS predicate. 39.04% data sources have provided information using both predicates. 34.28% of the data sources do not have working time details, or number of operating days per week. All of the 65.72% who have provided opening hour details also provided :open and :closes time. However no data source provided :validFrom or :validThrough opening hour specification. We also observed that 96.6% of the data sources have provided opening and closing times in UTC format and added ‘Z’ after time (e.g 10:10:10Z).
Making offering-related information available on the web in a structured format is one of the core objectives of GRO. The :Offering concept has 13 data properties that describe offering attributes and 16 object properties allowing several relationships to be established with other related concepts such as price specification, delivery options, payment or delivery charges, payment options, quantity and quality of products (included in offer and warranty). As previously stated, :Offering is the most widely used concept after :BuisnessEntity and is found in almost all eCommerce use case scenarios. Such as:
Find offering of a specific product and the available quantity Find delivery, warranty and payment charges of particular offering Responding to these question is dependent on the offering data landscape as illustrated in Table 7 and 8 with the different query and data patterns found. 31 It is important to note that herein, when we mention ‘Store’, it refers to the store, shop, branch office, office or any physical location, where the service or product is being provided on behalf of the store’s Company (:BusinessEntity)
From the perspective of data retrieval, useful information is available through relationships between different offerings and other related concepts. However, in order to either filter or restrict the search based on string matching, the textual description of the offer instance has to be relied upon. Another finding was that the prevalence of terms overlapping across vocabularies (such as :name and v:name), which also had to be considered when querying or generating customized rules. :availableAtOrFrom :acceptedPaymentMethods :includesObject :availableDeliveryMethods :hasPriceSpecification :includes Relationship patterns between offerings and other model concepts were looked at. The GR model provided two means of linking offering to products. When an offer has a single product, :includes is used, and for complex product bundling, :includesObject is used. In GRDS, 59.99% of data sources linked offers with product, and the remaining 40.01% use the offering concept to attach supplementary information such as eligible customer type, shop location information and supported payment methods. Tables 8 outlines the availability of relationships across the data sources with some not being used at all. 30.48% of companies provided price specification details, while 80.95% of the data sources identified the eligible customer type of the offer using GR predefined individuals such as :BusinessUser, :Endusers and :PublicInstitution. Within the GRDS, no data source provided information on the inventory level, advance booking requirement, delivery lead time, eligible duration of offer, eligible quantity to buy or eligible transaction volume. These omissions are not uncommon as this kind of information is required only for specific (unique) products not offered by web shops.
Consumers normally like to find offers containing some specific product and if found, they look for the product price, delivery method, payment details etc. The GR Primer36 mentions that at minimum, “the basic structure of an offering is always a graph that links (1) a business entity to (2) an offering. The offering itself is linked to one or multiple type and quantity nodes and one or more price specification nodes. Each type and quantity node holds the quantity. The unit of measurement for the quality, and the product or service that is included in the offering”. Retrieving offers with a specific product in mind would therefore require accessing concepts that relate product with offering and provide details on quantity and unit of measurement. As no formal product ontology is currently being exploited in the GRDS, we could only query offering and filter records based on a textual description attached to the offering. Tables 7 and 8 show an ‘offering’ data landscape. The following observations can be made: i) offerings can be retrieved with their price, quantity and offer start and end date; and ii) a filter clause can be applied to properties with literal values (such as :name, :description, rdfs:label and rdfs:comment) to narrow the search for specific offering.
GRO provides three means of describing products. Each approach has different structural requirements, allowing users to select the most appropriate. The first recommends using an appropriate product/service ontology to describe products referred to in an offering. The second and less structural approach allows lightweight product ontology tailored to individual specific needs using the GR top level Product or Service (:ProductOrService) concept and related vocabulary. The third, non-structural approach allows the description of product information lexically. This approach allows users to restrict their search to products with specific terms in their textual description. As in the previous use cases, we evaluated how GRDS responded to product related requests such as:
Result and Observations The query in Listing 3 was used to ‘ask’ the question which highlighted the lack of any formal product ontology to annotate products and their properties. We did however find that 2.86% of data sources used the second approach of using proprietary product ontology to describe quantitative properties. 97.14% of the data sources follow the third approach, publishing textual description of the product rather than the ontology. Two properties found in use for lexical information are rdfs:comment and rdfs:label. In general however we found no evidence within the GRDS of data sources using an appropriate product ontology. Since GR has aits core the description of offers, products can be searched either by exploring their offer data or through products included in the offers. In the absence of a proper product ontology we queries particular products by matching the keyword against the lexical information available in offer or product data.
Listing 3: Query (retrieving product description) The query in Listing 3 finds products containing “Cup” in their description, and displays price and associated currency, returning returned 58 products from three data sources37 with price and currency value.
RDFS and OWL define a set of forward chaining rules [11] which can be used to infer implicit knowledge and provide valid query results. The inclusion of implicit knowledge in the query result is 36 http://www.heppnetz.de/projects/goodrelations/primer/ 37 www.jarltech.de, www.overstock.com, www.corsetsandcurves.com.au achieved by using a reasoner with axiomatic triples available in an ontology. Despite the availability of customized rules for deductive reasoning, we focused on the axioms (listed in Table 3) available in GRO only. Based on ontological restrictions, the reasoning process helps to identify inconsistencies in instance data. We first looked at the instance data by applying the axiomatic triple using the RDFS rule set before using the class disjointness axioms, to perform disjointness checking in GRDS.
Implied information in the GRDS was investigated by applying an
axiomatic triple using the RDFS rule set to analyze the availability
of implied information at the data instance level. Using the RDFS
entailment rules [
In addition to subclass axioms, the GRO contains subproperty axioms which allow two resources, which are related through subproperty, to be implicitly related by superproperty. Figure 3 represents the subPropertyOf subsumption and transitive behaviour of data type properties. Of the 4 object properties recently added (2010-09-16), no instance data was found. As RDFS-style reasoning based upon backward chaining was used, it was necessary to rewrite queries so as to include implicit knowledge generated through RDFS entailment rules. In Figure 3(a) :hasCurrencyValue has two superproperties demonstrating that with RDFS-style reasoners any query with :hasCurrencyValue in its predicate will return three triples, two additional triples entailed by applying rule 7 and one original set of triples having the :hasCurrencyValue predicate. Figure 3(b) shows the results of applying reasoning over GRDS by using quantitative value concept’s data properties.
In web eCommerce, price data is often presented as a fixed price value and the user focuses on the :hasCurrencyValue property. Figure 4(a) draws attention to the fact that, apart from for one instance40, all data sources have provided only a fixed price value for their offerings. Data consumers will most likely use this property to access price value and, with the RDFS-style reasoner, its superproperties can return the same data. However, in a specific case where price range i.e. :hasMinCurrencyValue and :hasMaxCurrencyValue is provided and not :hasCurrencyValue, then using :hasCurrencyValue with or without reasoning will not return any value. Here, custom rules can be applied to return Max price value when there is no :hasCurrencyValue property value available. To handle similar kinds of situations, the GR website provides a set of GoodRelations Optional Axioms41 to allow users to obtain additional information from the dataset with minimal side-effects.
In Table 1, we saw that the disjoint class axioms in the GRO offer model consistency at the instance level. By making two classes disjoint, the same individual cannot be an instance of both (disjoint) classes simultaneously. For example, an individual declared to be an instance of class :Offering cannot be declared as an instance of :BusinessEntity since, in the GR model, both classes are defined as disjoint classes. The SPARQL query in Listing 4 finds such individuals and within GRDS we identified one data source42 violating the GR model. 38IF(uuu rdfs:subClassOf xxx AND vvv rdf:type uuu) THEN (vvv rdf:type xxx) 39IF(aaa rdfs:subPropertyOf bbb AND uuu aaa yyy) THEN (uuu bbb yyy) 40 http://plushbeautybar.com/services.html#PriceSpec_10 41 http://www.ebusiness
unibw.org/wiki/GoodRelationsOptionalAxiomsAndLinks 42 http://www.overstock.com/#company
Listing 4: SPARQL query In Figure 4, the same URI is used as an instance of type :BusinessEntity and :BusienssEntityType; whereas in the model, both classes are declared as disjoint classes.
A large amount of research work has been done on ontology
evaluation and a survey of different approaches is covered in [
In this paper, we analysed the implementation of the GRO by
consolidating 105 GR data sources into a single data set. We
analyzed the use of other ontologies with the GRO and
categorized data providers. Different use cases were used to
better understand and illustrate the schema usage and coverage
through ontological instantiation. Data sources provide structured
data aimed at improving search ranking only with no interlinking
currently available between eCommerce datasets or with LOD
[
Overall, the analysis points to early adoption and usage of an ontology that is beginning to achieve mainstream adoption with implementers using the GRO in an à la carte fashion rather than semantics a la mode.
In our future work, we intend to progress in two directions: i)
toward a more comprehensive analysis of an expanded dataset.
For this, we plan to collect datasets at intervals for a duration of
six months to determine whether the status quo remains
unchanged and, if not, how implementation develops with
increased maturity; ii) evaluating the usefulness of structured data
on the web. Here, we plan to investigate the impact of eCommerce
structured data (annotated using GRO and other eBusiness
ontologies) in search engine indexes (like Google, Yahoo!, etc)
and measure the increase in business activity, as has already been
evident in the traffic increase for BestBuy [
The authors would like to thank Michael Hausenblas and Aidan Hogan for useful discussion and advice. The work presented in this paper has been funded in part by the Science Foundation Ireland under Grant No. SFI/08/CE/I1380 (Lion-2) and the EU FP7 Activity ICT-5-4.3 under Grant Agreement No. 256975, LOD Around-The-Clock (LATC) Support Action and Activity ICT-4-2.2 under Grant Agreement No. 248458, Multilingual Ontologies for Networked Knowledge (MONNET).