Publishing Linked Data is a process that involves several design decisions and technologies. Although some initial guidelines have been already provided by Linked Data publishers, these are still far from covering all the steps that are necessary (from data source selection to publication) or giving enough details about all these steps, technologies, intermediate products, etc. Furthermore, given the variety of data sources from which Linked Data can be generated, we believe that it is possible to have a single and uni ed method for publishing Linked Data, but we should rely on di erent techniques, technologies and tools for particular datasets of a given domain. In this paper we present a general method for publishing Linked Data and the application of the method to cover di erent sources from di erent domains.
So far, Linked Data principles and practices are being adopted by an increasing number of data providers, getting as result a global data space on the Web containing hundreds of LOD datasets [3]. Moreover, Linked Data generation and publication does not follow a set of common and clear guidelines to scale out the generation and publication of Linked Data.
Furthermore, given the variety of data sources from which Linked Data can
be generated, we believe that it is possible to have a single and uni ed method for
publishing Linked Data, but we should rely on di erent techniques, technologies
and tools for particular datasets of a given domain. The rest of the paper is
organized as follows: Section 2 introduces our method for publishing linked data,
then, Section 3 describes the application of the method to cover di erent sources
from di erent domains, and nally, Section 4 presents the conclusions.
work we proposed a method that covers the following activities (
Application of the Method to Di erent Domains In this section we present the application of the method to cover di erent sources from di erent domains. 3.1
GeoLinkedData1 aims at enriching the Web of Data with Spanish geospatial data into the context of INSPIRE themes2. This initiative has started o by 1 http://geo.linkeddata.es/ 2 The INSPIRE Directive addresses 34 spatial data themes needed for environmental applications. http://inspire.jrc.ec.europa.eu/index.cfm/pageid/2/list/7 publishing diverse information sources belonging to the National Geographic Institute of Spain, IGN, and the National Statistic Institute in Spain, INE. Such sources are made available as RDF knowledge bases according to the Linked Data principles [3].
Regarding the speci cation activity, the data sources taken for this project
are composed of (
After the speci cation activity, we need to de ne the ontologies to be used
for modelling the domain of those data sources. We have created a modular
ontology network3 by reusing available knowledge resources. The ontology network
consist of the following subontologies (
As for the generation activity we have used (
Regarding the linking activity, we have used Silk12[1] for discovering owl:sameAs links between our dataset and external datasets, i.e., GeoNames13 and DBpedia14. We also used sameAs validator15, an application for validating the links discovered by Silk.
Once we had the legacy data transformed into RDF, we needed to store and publish the data in a triplestore. In our scenario we decided to rely on Virtuoso 3 http://geo.linkeddata.es/web/guest/modelos 4 http://vocab.deri.ie/scovo 5 http://www.fao.org/countryprofiles/geoinfo.asp 6 http://www.oeg-upm.net/index.php/es/ontologies/107-hydrontology 7 http://www.w3.org/2003/01/geo/ 8 http://www.w3.org/TR/owl-time/ 9 http://www.oeg-upm.net/index.php/en/technologies/9-r2o-odempaster 10 http://www.oeg-upm.net/index.php/en/technologies/57-nor2o 11 http://www.oeg-upm.net/index.php/en/technologies/151-geometry2rdf 12 http://www4.wiwiss.fu-berlin.de/bizer/silk/ 13 http://geonames.org/ 14 http://dbpedia.org/About 15 http://oegdev.dia.fi.upm.es:8080/sameAs/ Universal Server16 and Pubby17. For the metadata information we have used VoID18 for expressing the metadata about our dataset.
Finally, for the exploitation we have developed a prototype application19 that browses and visualizes the RDF data. 3.2
In AEMETLinkedData20 we have generated Linked Data from Spanish meteorological data. Within this initiative we are publishing information resources from the Agencia Estatal de Meteorolog a (Spanish Meteorological O ce, AEMET), as Linked Data.
Within the speci cation activity, among all of the data made available in the FTP server from AEMET, we have focused on surface meteorological observing stations, and more precisely in measurements taken in ten minute interval times. AEMET has around 250 automatic weather stations of this network, registering pressure, temperature, humidity, precipitation and wind data every 10 minutes. Data from the di erent stations are publicly available online in comma separated values (CSV) les, compressed with gzip, updated every hour and kept for seven days. This means, every hour six new les are added, corresponding with periods of ten minutes, and every day a new folder is created to store the les for that day.
As for the modelling, the development of the AEMET ontology network21 has
been performed following an iterative approach based on the reuse of existing
knowledge resources, both ontological (including ontologies and Ontology Design
Patterns ) and non-ontological resources as proposed by the NeOn methodology
[5]. The AEMET ontology network follows a modular structure consisting of a
central ontology that links together a set of subontologies that describe di
erent sub domains involved in the modelling of the meteorological measurements.
The subontologies are (
The RDF generation is produced with ad-hoc Python scripts25. These scripts are executed in two processing steps, integrating with ease the generation of RDF and tasks as crawling the FTP server where the CSV les are located. The rst step generates the information about the automatic stations, which is static, and thus needs to be executed only once. The second step generates the information about the observations on a regular basis, keeping the data updated.
Regarding the publication, the generated RDF is stored in Virtuoso, which integrates with Pubby for the publication of the results and makes them available for humans as well as computers.
As for the exploitation, we have developed a simple application26 that consists on a viewer based on a modi ed version of map4rdf27, a tool that uses Google Web Toolkit framework to visualize and access Linked Data resources. Each automatic weather station is geolocated, enabling its representation in the map with a marker. Since there are approximately 300 stations, they can be all represented at the same time. 3.3
El Viajero is a dataset that exposes descriptive metadata and provenance of news, blogs and posts belonging to the Prisa Digital Group28 in the domain of travel journalism. El Viajero makes use of several heterogeneous datasets, and it is integrated within a service which exploits it in order to help users navigating and browsing the contents of the dataset.
Regarding the speci cation, the data sources analyzed come from a variety of newspapers and digital platforms belonging to the Prisa Digital Group: \Suplemento El Pa s", \Gu as Aguilar", and \Canal Viajar". Users can also create their own blogs to post their traveling experiences (around 600 di erent blogs exist), apart from commenting the news posted by the reporters of the group. The news are stored using an extension over the International Press Communications Council (IPTC)29 Standard called News Industry Text Format (NITF)30, which is also used by other major publishing agencies in Europe, such as AFP31, ANSA32 or AP Digital33. The data is separated from the metadata in di erent kinds of les: one for the structure of the news, another for the content of each piece of news, another one for the metadata of the sources of the news like images or videos and another one for storing statistics about the news (such as views, date of the last view, etc). The blogs and posts are stored in two di erent private databases: one with metadata about the posts (creator, number of comments, 25 http://aemet.linkeddata.es/technology_en.html 26 http://aemet.linkeddata.es/browser_en.html 27 http://oegdev.dia.fi.upm.es/map4rdf/ 28 http://www.prisa.com 29 http://www.iptc.org/site/Home/ 30 http://www.iptc.org/cms/site/index.html?channel=CH0107 31 http://www.afp.com/ 32 http://www.ansa.it/ 33 http://www.apdigitalnews.com/ body, date of creation, etc.) and the other one with metadata about the blogs (title, header image, description, owner, etc).
The resources have been modeled following a layered approach. In the rst level DC34, SIOC35, MPEG736 and WGS8437 vocabularies are used for describing the metadata of the di erent resources on the platform (e.g., post metadata, photo and video metadata, location metadata, etc.), but not their evolution. In the second level, our ontology38 connects the previous vocabularies to the Open Provenance Model (OPM) [4] by extending the OPM core ontology (OPM-O39) in the third layer.
As for the generation activity, we have developed and reused parsers to produce RDF from the Prisa Digital datasets automatically. The system has been developed using the JENA Framework, combined with the TDB and Virtuoso repositories for the proper publication of the dataset. The data is exposed using Pubby, and additional metadata about the query used to retrieve the exposed data is captured with the Provenance Vocabulary (along with metadata of the dataset itself).
Since El Viajero belongs to a Spanish media company, many of the guides0 locations refer to Spanish locations. GeoLinkedData contains accurate metadata about the majority of these locations, so we have linked to this dataset the guides within the Spanish territory. For the guides referring to the rest of the world we linked to DBpedia, the most linked dataset in the Linked Data cloud, and which provides many additional details about the locations. We relied on the Silk framework for nding the links to these two datasets, curating manually the results obtained by the tool. Additional links from travel recommendations have been linked to the Linked User Feedback (LUF) dataset40.
Thanks to the publication of El Viajero as Linked Data, users have available more than 6600 resources about almost 1000 locations around the world. In order to help them organize, browse and explore the contents of the dataset, we have developed an application [2] to expose these contents in an eye catching way by representing them in a map. 3.4
datos.bne.es datos.bne.es publishes bibliographic data from the Spanish National Library (BNE) catalogue. This initiative has kicked o with the publication, under Linked Data principles, of information from the bibliographic and authorities catalogues, making them available as RDF knowledge bases. Furthermore, these bases are interrelated with other knowledge bases existing within the Linking 34 http://dublincore.org/documents/dcmi-terms/ 35 http://rdfs.org/sioc/spec/ 36 http://metadata.net/mpeg7/ 37 http://www.w3.org/2003/01/geo/ 38 http://webenemasuno.linkeddata.es/Ontology/OWLDOC/index.html 39 http://openprovenance.org/model/opmo 40 http://lab.isoco.net/technologies/luf Open Data initiative. With this initiative, Spain has joined the ranks of other institutions such as the British Library and the Deutsche Nationalbibliothek that have recently launched similar projects.
Regarding the speci cation activity the data source taken for this project is composed of around 3.9 million bibliographic and 4.2 million authority records in the MARC2141 format using the ISO 270942 encoding standard, provided as data dump from the BNE catalogue. With the respect to the data quality, we have assumed the records to present necessary quality features like consistency, accuracy, or precision, as the records have been created and curated by highly quali ed professionals. Regarding the URI design we have put special focus on minting URIs in a multilingual scenario, taking into account the design of TBox and ABox URIs.
After the speci cation activity, we de ne the ontologies to be used for modelling the domain of those data sources. Again, the most important recommendation is to reuse as much as possible available vocabularies and ontologies. To this end, we have decided to reuse a very well stablished set of models as the IFLA Functional Requirements family43 and ISBD elements44 that have been agreed upon and used by the library community.
As for the generation activity we have used MARiMbA45 that provides a tool focused on enabling and easing the mapping from MARC21 to the RDF(S)/OWL models, and the transformation into RDF. In this activity we have emphasized the importance of the mapping in which we establish correspondences between MAR21 records the chosen vocabularies. Furthermore, MARiMbA0s main design principle is to provide an easy-to-use mapping framework that allows domain experts (i.e. librarians and cataloguers) to work independently from the IT team and to establish the mapping rules without the need of high technical skills (e.g. knowledge about XML, XSLT or other ad-hoc mapping languages).
Regarding the linking activity, we chose VIAF46, the Virtual International Authority File from the OCLC, as a candidate dataset since we had to deal with catalogue authoring data. VIAF provides a cluster of authority records of the same entity across several national authority les. This cluster has proven to be useful to establish links to other authoritative datasets such as the German National Library.
Once we have the legacy data transformed into RDF, we need to store and publish the data in a triplestore. In our scenario we have decided to rely on Virtuoso Universal Server and Pubby and to provide an additional web-developersfriendly API (providing simple methods for retrieving persons, works, etc.), using 41 http://www.loc.gov/marc/bibliographic/ 42 http://www.iso.org/iso/iso_catalogue/catalogue_tc/catalogue_detail.htm? csnumber=7675 43 http://www.ifla.org/en/node/2016 44 http://metadataregistry.org/schema/show/id/25.html 45 www.oeg-upm.net/index.php/en/downloads/228-marimba 46 http://viaf.org/ Puelia47. For the metadata information we have used VoID for expressing the metadata about our dataset.
Finally, for the exploitation we have developed a prototype application that provides easy full-text searching and navigation based on the FRBR model. 4
Conclusions In this paper we claim that it is possible to have a single and uni ed method for publishing Linked Data, but we should rely on di erent techniques, technologies and tools for particular datasets of a given domain. Therefore, we have presented our method for publishing Linked Data and the application of that method to di erent domains. Table 1 summarizes the vocabularies, technologies, and applications used in the development of our initiatives. The rst column represents the six main activities of our methodological guidelines, i.e., modeling, RDF generation, links generation, publication and exploitation. The rest of columns include the technological support used for each phase within the set of our initiatives. 47 Puelia is an implementation of the Linked Data API speci cation, see http://code.google.com/p/puelia-php/
Moreover, we have to take into account that there are (
This work has been supported by BabeLData (TIN2010-17550), myBigData (TIN2010-17060), PlanetData (FP7-257641), and Webn+1 (TSI-020301-200924) projects. We would like to thanks to all OEG members involved in the Linked Data initiatives.