2.1. Linked Data

Tim Berners-Lee initially proposed the principle of Linked Data in 2006 as a practical approach to the implementation of what the Semantic Web had envisioned as a collaborative effort between human beings and machines in the exploration and discovery of useful data and new connections (Berners-Lee, 2006). Technically, LD is not only clearly defined and machine-readable but also incorporative of mechanisms oriented toward linkage to external data. LD is also defined as openly published data that can be referenced by external datasets (Bizer, Heath, & Berners-Lee, 2009). Whereas the traditional Web of Documents simply connected HTML documents via hyperlinks, the new Web of Data provides rich connections between raw data using RDF, resulting in information that is richer in meaning and more open to the creative data processing of human and non-human users than ever before. With such advantages of Linked Data in mind, Berners-Lee recommended that LD be published according to the following four principles: 1) Use URIs as names for things; 2) Use HTTP URIs so that people can look up those names; 3) When someone looks up a URI, provide useful information using standards (RDF*, SPARQL); and 4) Include links to other URIs so that they can discover more things (Berners-Lee, 2006).

The main characteristic of LD lies in its potential to create valuable knowledge based on diverse data by sharing identifiers. The LD tends to be generated in various sizes and domains by individuals and institutions. The distributed nature of LD expansion poses serious challenges in enhancing LD quality since the former depends on the degree of connectivity based on common identifiers. Even though we have many well-established identifiers at both class and property and instance levels, the actual deployment of them has been lacking in a systematic manner.

2.2. LLD (Library Linked Data)

LLD (Library Linked Data) refers to a dataset published on the Web in accordance with the four LD principles recommended by Tim Berners-Lee. Library data is typically made available in the MARC21 or MODS formats. This data is usually converted into LD through database mapping or through conversion modules. Assigning unique identifiers in the form of Uniform Resource Identifiers (URIs) is the core of Linked Data. As library data already has standard identifiers and links among items based on subject authority, it has the advantage of being ready for publication as LD of higher quality than other datasets. The clear identification systems and cross-domain subjects of LLD make library data much more valuable when integrated with other datasets. In order to integrate LLD with the global LD, we need a smart search system that can help one to find identifiers associated with classes, properties, and instances of the worldwide LD. It is crucial that this system is regularly updated to provide seamless linking to the existing data.

2.3. Open Data

Much like open source or open access, open data is data free from copyright, patents, and similar control systems; it is available to anyone who wishes to access and republish it (Auer et al., 2007). Although the concept of open data is not new, its popularity began with the advent of the Web. In recent years, the United States, United Kingdom, and major research institutions (including government agencies and businesses) have chosen to open their data to the public. The amount of data being published on the Web has been increasing exponentially each year. Most of the data made open via the Web are available in the formats of CSV, Excel, plain text, or XML/ JSON via OpenAPI. These kinds of open data can be easily converted into LD, enabling them to be integrated into other data since its data formats are compatible.

3.1. Datahub.io (http://datahub.io/)

Datahub is a data registry service launched in 2007. It was developed using the CKAN (Comprehensive Knowledge Archiving Network) open platform maintained by the organization OK (Open Knowledge). CKAN is a Web-based open-source data management system for data repository. Datahub provides information on diverse datasets and details on licensing, dataset sizes, data owners, and SPARQL endpoints. Datahub is used as government data catalogs for UK’s data.gov.uk and US’s data.gov services. Open Knowledge, formerly known as the Open Knowledge Foundation, promotes open data. This site provides about 10,000 datasets and search services to them via diverse conditions. For example, searching can be done by a publishing entity, data format, usage condition, and keyword. However, it is very difficult to find detailed information regarding each dataset and its relationship to others.

3.2. LODStats (http://stats.lod2.eu/)

LODStats tracks the structure, coverage, and coherence of datasets on the Web (Demter, Auer, Martin, & Lehmann, 2012). LODStats gathers statistics about datasets related to the Resource Description Framework (LODStats, 2014). LODStats is comprised of a LODStats core module that can handle diverse LOD requests and a frontend module capable of providing the user with the Web interfaces, search classes, and data properties contained in LOD. Ermilov (2013) has developed a LODStats Web application that makes it possible to gather comprehensive statistics on LOD status and investigate the usage of vocabularies, classes, and properties. The LODStats system can be used to monitor real time changes of large CKAN data. Figure 1 shows the Python-based architecture of CKAN (Ermilov, Martin, Lehmann, & Auer, 2013).

Fig. 1

LODStats system architecture

As can be seen in Figure 1, LODStats holds major components needed for LLD repositories. It provides a SPARQL module to support queries for LD dump and a related module to confirm any updated status of data stored in the CKAN platform. LODStats also supports a search mechanism to find detailed information regarding classes and properties of each dataset.

3.3. Datacatalogs.org (http://datacatalogs.org/)

DataCatalogs.org is the most comprehensive list of open data catalogs in the world. The purpose of Data-catalogs.org is to provide users with a list of trustworthy ODR (open data repositories) selected by open data experts from all over the world. The first version of Datacatalogs.org was released at OKCon 2011 in Berlin, Germany. As of November 2014, the list includes detailed information on approximately 390 ODR. The strength of this service is its association with maps so one can easily find who is running each repository and its regional information, which is usually not available at most repositories.

3.4. European Union Open Data Portal (https://open-data.europa.eu/en/data/)

The European Union Open Data Portal is the main access point for data from institutions and other organizations of the European Union (EU). The EU Open Data Portal consists of a catalog of major EU and EU-related organizations that provide data in LD format. Users can take advantage of this data for either commercial or non-commercial purposes. SPARQL endpoints may be used to search the data. As of November 2014, more than 7,600 linked datasets have been made available through the EU Open Data Portal.

4.1. Introducing LOV: http://lov.okfn.org/dataset/lov/

Linked Open Vocabularies (LOV) is a portal for RDFS and OWL ontologies developed by Bernard Vatant and Pierre-Yves Vandenbussche as a crucial part of the framework for the Datalift (http://datalift.org) project (Scharffe et al., 2012). The service has constantly been updated and refined and was selected as an official OKF project in 2012. LOV aims to achieve the same purposes as most ODR, furnishing users with a visual rendering of how LOD is used in Linked Data cloud settings and helping them design their own ontologies using available classes and properties. LOV uses the terms “vocabulary” and “ontology” interchangeably, with an emphasis on reusability, integration, and association.

The following are core criteria for inclusion in LOV:

LOV provides detailed descriptions of ontologies formulated in RDFS or OWL, both of which are widely used in Linked Data. The following are the characteristics required for an ontology to be included in LOV:

According to 2011 criteria, an (ideal) appropriate data size is defined as 10 classes and 20 properties. More than 80% of LOV vocabularies have less than 100 elements. The largest LOV vocabulary is schema.org, with more than 500 elements.

The requirements presented above could render LOV more advantageous for the library community than other open data systems. LOV not only provides the user with diverse search capabilities concerning registered vocabularies but also focuses on datasets freely available under Creative Commons licenses for all purposes (including commercial usage). The following is the description of the four functions provided by the LOV system.

Table 1.

The Functions Provided by LOV

4.2. LOV System Analysis

The LOV system architecture is shown in Figure 2. The modules are integrated and work as a single Web application.

Fig. 2

LOV architecture

The functions of the main modules in the LOV architecture are shown in Table 2.

Table 2.

Module Functions of LOV System

All LOV modules are implemented using Java language. The LOV system provides search API calls which can easily be interfaced with external systems. When a search API call is made to find elements or ontologies using the HTTP GET format, result values are returned in JSON format. The core of LOV is composed of Bot and Aggregator modules, which regularly update LOV for the latest versions of registered ontologies using SPARQL rules. The LOV modules also provide all statistics regarding updated ontologies relevant to LD disciplines.

4.3. Analyzing LOV Data

As of December 2014, LOV provides information regarding 10 categories and 421 ontologies. Table 3 is a summary of the sub-categories and number of ontologies that belong to each of these 10 categories. The major 10 categories are further divided to help users easily find representative ontologies in a given subject area.

Table 3.

LOV Ontology by Category

The categories with the largest number of ontologies are ‘Data & System’ and ‘Library,’ both of which are solid references to consult when implementing LLDs. The ‘Science’ ontology can be used when publishing LLD in the natural sciences, just as the ‘Media’ ontology is applicable to LLD on library multimedia resources. As of 2011, the vocabularies tracked by LOV have shown a high reusability rate, resulting in LOV providing quality data and sufficient mechanisms for linkage with other data (Poveda Villalón, Suárez-Figueroa, & Gómez-Pérez, 2012).

4.4. LOV Comparison

As shown in Table 4, Braunschweig (2012) has compared diverse open data repositories published by various governments across the world and proposed nine criteria for repository evaluation. The criteria are intended to measure data size, data quality, and degree of usability.

Table 4.

Comparison Element of Data Repository (Braunschweig, 2012)

The results of the LOV data investigation using the criteria depicted in Table 4 are available at http://wwwdb.inf.tu-dresden.de/opendatasurvey. Table 5 shows the results of applying the nine criteria developed by Braunschweig (2012) to LOV data and other systems mentioned in previous studies for the purpose of testing the overall usefulness of the LOV system.

Table 5.

Comparison of LOV and Major Open Data Repositories

As can be seen in Table 5, LOV and the EU Open Data Portal satisfy most of the criteria specified above. While the coverage of LOV is relatively small compared to other repositories, this is due to the strictness of the criteria applied before any institution can register data to LOV. Due to the meticulous maintenance of LOV ontology quality, LOV is able to provide better quality data and exhibit a wider range of applicability compared to other open data repositories. The major strength of LOV is in its ability to specify how data are connected among different ontologies. The LOV provides visualization of relationships among internal and external ontologies when one searches for a particular vocabulary. In addition, the LOV provides all the versioning information regarding before or after ontology updates in the system, which is not offered by other systems.

5.1. Implementing an Open and Specialized LLD Repository

Studies on and examples of LD and repository construction have tended to focus on government agencies and industrial sectors. Similar case studies in the library community are comparatively lacking. While a few major libraries have attempted to issue and publish LLD on the Web, these instances were made possible by the efforts of individual institutions rather than a collective trend. With the construction of various LD-based application services underway, an LLD repository system will be required for the library community to be able to conduct sophisticated searches and filter for data with greater ease. Trained professionals must implement LLD by actively utilizing standard identifiers, thereby increasing the quality of LD and making the linkage of library data to other open repositories more meaningful.

From this perspective, LOV can be used to build LLD-specialized repositories. Based on our analysis of the LOV system, LOV is equipped with all the functionalities needed to run an LLD repository. In addition, the LOV open source codes can be modified to maximize the repositories for the needs of the library community.

5.2. Applying LOV to Library Systems

Most libraries use commercial or open-source data management systems to organize their bibliographic and authority metadata. With the need to publish bibliographic data as LD on the rise, libraries should be prepared to publish LLD and integrate LLD with external LD. One approach is to implement an LLD-specialized repository system. The LLD-specialized repository system can enable the library community to query or import its data into library data, with LOV being utilized as a separate repository associated with a library system or as a module with the system. In other words, data mapping may be achieved by searching for LLD-specialized repositories to ingest metadata into a library system. Should this be accomplished, library systems would be able to build models with the capacity to publish data as LLD and automatically integrate said LLD into repositories. Such real-time ingestion from and publishing into repositories would in turn maintain the efficiency of LD publishing updates.

5.3. Applying LOV to Big Data Archiving

Recently, both academia and the industry sector have used Big Data analytics to study and derive meaningful results from massive amounts of data (i.e., social data or internal data). Traditionally, ontologies have been widely used for mapping, identifying, and integrating meaning in data analysis—a role expected to extend to big data analytics. As libraries have extended our knowledge of data collection and preservation, it is likely that they will play an important role in big data management. With such anticipations in mind, the LOV system presents the promising possibility of classifying, mapping, and analyzing data to be collected and preserved. LOV can be utilized in the analysis of mapping data types, major concepts, and data elements. In addition, LOV can be useful in enhancing the quality of LD and LLD by creating previously nonexistent associations. In order to produce continuous values from big data, we should be able to feed refined data based on the criteria from diverse contexts into big data analytics. In an effort to do so, discerning which data or ontology needs to be consulted for big data analysis at different contexts will be important and LOV could be employed meaningfully in this step. When archived big data linked with LOV are fully utilized, it will add value to big data and provide an incentive for archiving them.