Article Preview
Top1. Introduction
The “Linked Data approach” is the label for a set of techniques of publishing structured data on-line so that it can be interlinked and become more useful. Currently, Linked Data relies on standard Web technologies as the Hypertext Transfer Protocol1 (HTTP), to retrieve data from the World Wide Web2 (WWW) and the mechanisms of Uniform Resource Identifiers3 (URI) for identification. It combines this with a rich representation of datasets based on Resource Description Framework4 (RDF). The Web of Linked Data can thus be defined as a way to share open and structured data on the Web. Based on four principles exposed by Tim Berners-Lee in 20065, the objective is that of making the data more readable to computers by publishing it in RDF and using URIs to identify each resource. Finally, data can be retrieved using the standardized language and protocol SPARQL6 and RDF Query Language. These approaches can be framed in the field of the Semantic Web, (Albert, Berners-Lee, & Ficheti, 1999) or (Berners-Lee, Hendler, & Lassila, 1999). A general overview of what Linked Data is can be found in (Bizer, Heath & Berners-Lee, 2009).
The adoption of Linked Data best practices, described in (Heath & Bizer, 2011), has resulted in the publication of datasets by data providers in several domains. Datasets use RDF as language and expose triples to describe the information. Each triple has the structure of subject, predicate and object. Subjects are URIs representing resources. Objects could be URIs or particular values. Predicates are also represented by URIs and are the way to relate a subject with its object. The entities represented by the URIs can be looked up by using the HTTP protocol. In the case that the subject and the object of a triple belong to different datasets a link between them is created, it is what is often called a RDF link. Taking into account the different datasets and its RDF links, the Web of Linked Data (LOD) can be seen as a graph. Here the datasets will represent the nodes and the RDF links the edges between them. This structure of graph allows applications to navigate between them and discover new information. By adding new datasets, the Web of Linked Data will evolve changing its structure. Since 2007, it has increased its size from a dozen of datasets to more than a thousand. A recent review can be found in (Schmachtenberg, Bizer, & Paulheim, 2014).
The techniques used to analyse the Web as a graph can also be applied to the Web of Linked Data. By studying the link structure of LOD, different measures used in Network Analysis can be applied, (Wasserman & Faust,1994): how to crawl information from it, importance of datasets by their sizes or understanding its behaviour looking at the evolution.
This paper is reporting on the main empirical findings on a comprehensive analysis of the graph structure of the Web of Linked Data. Following approaches used successfully in the past to analyse that structure for the Web of documents. This led to important insights helping to innovation in data retrieval strategies or an increased understanding of the social structure of Web communities. First of all, some metrics about the datasets in LOD and its usage are obtained. Then Social Network Analysis (SNA) techniques are applied, providing a general picture of the Web of Linked Data.
Results show that the structure of the Web is very compact which a low distance between nodes. The nodes also have a reasonable number of edges and most of them are close to the rest of the graph. Finally, it can be concluded that the structure of the Web fits with the theory of the bow-tie.