pCaoppeyrrsightby© 2th0e10 fpoarpethrse' inaduivthidoursa.l Copying permitted only for private and aepdcuaibtdloisershm.eidc paunrdposecos.pyTrihgihstedvolubmye itiss
10-Jun-2010: submitted by Christoph Lange 11-Jun-2010: published on CEUR-WS.org
Ontology Recommendation for the Data
Publishers?
Antoine Zimmermann1
Digital Enterprise Research Institute National University of Ireland, Galway, Ireland
firstname.lastname@deri.org 1
RDF data publishing on the Web has gathered momentum in the last few years, thanks to a general e ort to link open data all over the Web. Yet, this trend is slowed down by the di culty to nd appropriate terms for the data to be described. Indeed, apart from a handful of well known ontologies, there are no readily available, easily ndable vocabularies for most of the domains that would be good candidates for publishing data in a standard, linkable way.
The typical problems faced by would-be data publishers are: (1) ontologies de ning the domain of interest do not exist; (2) they exist but are di cult to nd because developed by small groups for experimentation, lacking advertisement; (3) they exist and can be found but they are of poor quality, not complying with standards or best practices; (4) they exist and can be found but there are too many, of mixed quality, and it is di cult to assess which ones are appropriate for a speci c use case.
To address the rst issue, user-friendly ontology editors have been developed Swoop1, Protege2, etc. But this mostly requires that ontology and domain experts publish more and more data and terminologies. We assume that this will naturally happen when Linked Data and Semantic Web technologies will reach a critical mass which trigger a virtuous circle. We will not address this issue here.
The second item is somehow addressed by Semantic Web search engines.
Several of them have been proposed, such as Swoogle [
Therefore, we believe that to guarantee access to the ontologies that are available, well supported and of quality, there is a need for promoting them more actively. In this paper, we argue in favour of having a committee of experts analyse Web vocabularies with respect to their suitability as reusable terminologies for data publishers. As a result of this analysis and evaluation|which can be partly automatised|the committee would advertise the ontology as a \quality vocabulary" and recommend it for describing information in the eld applicable to such terminology.
To present this idea in more details, we rst discuss the criteria that a Web terminology should ful l to be labelled as \quality vocabulary" (Section 2). Then, we describe a possible approach to implement such an evaluation and recommendation framework (Section 3). Finally, we show how this integrates with ontology repositories (Section 4). 2
Our objective in recommending Web vocabularies is focused on helping data publishers to nd adequate terms for describing their data. For this reason our proposed initiative distinguishes itself from other ontology evaluation activities that focus more on engineering, design and logical issues. Moreover, we do not pretend to assess the quality of the modelling of the domain of interest, which could only be judged by a domain expert. Also, we encourage small, lightweight ontologies, which are easier to assess, reuse and scale. In this section we discuss possible criteria for quality vocabularies, recommended for reuse over the Web of Data. More precisely, data publishers would expect vocabularies that are: (1) justi ed by use cases; (2) easy to reuse and publish; (3) well interoperable with published Linked Data. We analyse these requirements to determine the criteria for quality vocabularies.
Justifying the existence of the vocabulary. As a primary requirement, a vocabulary should be accompanied by a statement about its utility. This includes a general description of the vocabulary and its scope as well as, more importantly, its related use cases. To avoid too much subjectivity in deciding the relevance of a vocabulary in terms of usage, it can be required that a Web vocabulary proposed for recommendation should be supported by at least some data publishers. We consider this criteria a very important one and would not recommend a vocabulary, be it very well designed, if nobody considers using it. Usage should not be restricted to a unique dataset, not even to a big one by a major player in the eld. At least two independent parties should be using the terms, or there should be strong evidence that the terms will be used by several distinct publishers in the near future. As an alternative proof of relevance, the vocabulary publisher could claim potential adoption by showing precise examples of possible usage. E.g., a geo-location vocabulary can be proven to be useful if the authors show that translating existing geographic databases into linked data can be done in an easy and straightforward way to create and publish multiple datasets at a low cost. Finally, the utility of the vocabulary can be demonstrated if existing applications are usefully exploiting the associated data. This can justify the recommendation of a vocabulary since all data conforming to it will interoperate with those existing applications.
Ease of reuse and publication. Since one of the goal of recommending
vocabularies is to increase interoperability by reducing the number of heterogeneous
terminologies, it is important that the vocabulary be reusable as easily as
possible. To achieve this, the content should be made understandable by non-ontology
experts. Thus, one of the criteria is the presence of clear labels and textual
descriptions for each term in the ontology. Moreover, granularity and complexity
should be in line with the use cases. Highly expressive or too speci c
ontologies should be discouraged if they are not seriously justi ed. Finally, publication
of data conforming to the proposed vocabulary should be made easier, e.g., by
providing tools that automatise (partly or totally) the publishing process. For
instance, FOAF and SIOC exporters make the creation of online community RDF
metadata fully automatic when integrated in a content management system.
Interoperability. To ensure better interoperability, several guidelines have to
be followed. Obviously, vocabularies should be published in a standard format,
namely RDF(S) and OWL. Additionally, since vocabularies are themselves part
of the Linked Data, they should follow the best practices in the eld [
This section shows how a framework for recommending Web vocabularies could be implemented in practice.
We notice that most ontologies and Web vocabularies, especially the most popular ones, are developed by academic researchers (FOAF, SIOC, Good Relations, Music Ontology, etc.) Thus, it would be possible to incite the ontology builders to publish their creation through our quality assessment process by establishing a regular submit/review/accept-reject process. An ontology or terminology for data publishing o ers a solution to a problem or ful l a certain need. It can thus be seen as a scienti c contribution. We propose to have a call for vocabularies with a review process and editorial constraints.
First, an automatic tool will verify the compliance of the submitted vocabularies to the well established criteria mentioned in Section 2. If these are matched, then a peer-reviewing process will be undertaken by the committee, considering what has been discussed in the previous section. A submitted ontology must be accompanied by a description that will be published together with the ontology. The description|which can take the form of an article|must explains the purpose of the ontology|not only its domain but also its scope and granularity as well as possible or existing applications using it. It must show the utility and the need for such a vocabulary. This is partly proven by the fact that existing (independent) datasets are already using the terms or publishers have committed to use it in the near future.
The descriptions should be kept understandable by non-ontology specialists, and technical details can be given if, and only if, it contributes to showing the utility and interoperability of the vocabulary. As a result of this process, the ontology is either deemed not suitable for recommendation or recommended as a \quality vocabulary". Rejected ontologies can be improved and resubmitted later, eventually leading to better quality of the vocabularies. A centralised Web site would advertise these ontologies, provide documentation about them and allow searching and browsing them. Such a central place could be an existing ontology repository, as discussed in the next section. 4
The recommended ontologies should be easily searchable, browsable as well as matchable. These are common tasks made possible by ontology repositories. Therefore, the recommendation process that we described previously may be integrated into a repository that could possibly include other non-recommended ontologies. However, the recommended ones should be emphasised and all operation should be applicable to the quality ontologies only. Moreover, non-recommended vocabularies could be marked by speci c labels indicating that, although they are not recommended, they validate some of the criteria for quality. 5
In this paper, we argued that data publishers should be guided in their choice of vocabularies by actively recommending them ontologies that are considered of quality. The recommendation should be based on criteria that span from support by existing publishers and applications to the compliance to the best practices in this eld. Evaluating those criteria could take the same form as an academic publication process, with a review phase. As a rst step, we would like to launch a workshop on this topic which would create, we hope, an incentive for researchers to design and publish quality ontologies for the Web of Data.