A Fuzzy RDF Graph-Matching Method Based on Neighborhood Similarity

Introduction

Resource description framework (RDF) (Klyne & Carroll, 2006) is a standard data model recommended by the World Wide Web Consortium (W3C) to capture resource information the context of the semantic web (Berners-Lee, Hendler, & Lassila, 2001). This model represents data as sets of triples where each triple consists of three elements that are referred to as the subject, the predicate, and the object of the triple. These triples allow users to describe arbitrary things in terms of their attributes and their relationships to other things. At the same time, information is often vague or imprecise in the real-world applications. Therefore, the study of fuzzy extension of RDF models has emerged (Ma, Li, & Yan, 2018; Ma, & Yan, 2018; Mazzieri, & Dragoni, 2008; Straccia, 2009). Nowadays fuzzy RDF has been widely used in a variety of real scenarios (Pivert, Slama, & Thion, 2016; Zhang, 2017).

An increasing amount of data is becoming available in RDF format. Multiple datasets are effectively published according to the linked-data principles (Zhang, Song, He, Shi, & Dong, 2012). Integrating these datasets through interlink or fusion is needed in order to assure interoperability between the resources composing them. To do this, we need to automatically discover the correspondence between these data sources in different information stores. Data matching is the process of bringing data from different data sources together and comparing them in order to find out whether they represent the same real-world object in a given domain (Dorneles, Gonçalves, & dos Santos Mello, 2011). Efficient RDF data matching becomes the technical foundation of many tasks in Semantic Web (Li, & Ma, 2018; Zou, & Özsu, 2017).

Fuzzy RDF data have a natural representation in the form of a labeled directed graph in which the vertices present the resources and values (also called literals), and edges represent semantic relationships between resources. So, RDF data matching problem has been often addressed in terms of graph matching approach. Graph matching is usually based on the graph isomorphism or homomorphism, combining a specific application environment to find similar topology graph. Some works (Carroll, 2002) has been dedicated to the search for the best match between two graphs or subgraphs. Unfortunately, the traditional graph matching algorithms based on graph isomorphic have been proved that its complexity is NP-complete (Ullmann, 1976). For this reason, some works (Costabello, 2014; Dorneles, Gonçalves, & Mello, 2011) of approximate matching based on similarity or distance metrics use a specific index structure to reduce the complexity of RDF graph matching. However, these approximate matching approaches ignored many features of RDF graph. Firstly, these approaches only take the similarity of vertices and edges into account in RDF graph, do not concern structure among the vertices and edges. Secondly, vertices in RDF graphs have incomplete information or even anonymized information (blank vertices), and the partial neighborhood information available from a source graph will be helpful to identify entities in the target graph. More importantly, these methods cannot process fuzzy information in the matching process.