Linked Open Data (LOD), a powerful mechanism for linking di erent datasets published on the World Wide Web, is expected to increase the value of data through mashups of various datasets on the Web. One of the important requirements for LOD is to be able to nd a path of resources connecting two given classes. Because each class contains many instances, inspecting all of the paths or combinations of the instances results in an explosive increase of computational complexity. To solve this problem, we have proposed an e cient method that obtains and prioritizes a comprehensive set of connections over resources by traversing class{class relationships of interest. Based on the method, we developed a system for constructing a SPARQL query named SPARQL Builder. We showcase how to generate a SPARQL query according to user's interest by using the SPARQL Builder system.
(resources) whose types are given two classes for integrative data analysis with semantics. These paths can be obtained by retrieving chains of properties (links) which connect instances of classes. In other words, these paths can be obtained by traversing paths of class{class relationships over the LOD.
Therefore, based on class{class relationships, we have been developping a system named SPARQL Builder to obtain data from LOD exibly, by assisting users in writing SPARQL queries to the SPARQL endpoints. To realize our approach, we should develop the following two techniques: 1) a method to collect pro les related to class{class relations through SPARQL endpoints of RDF datasets: This is implemented as SPARQL Builder Metadata (SBM), which describes comprehensive metadata including not only class de nitions but also statistics such as the number of instances while it is not supported existing metadata. 2) a method to obtain chains of properties and classes by computing paths on labeled multigraph named class graph: This enables an e cient method to compute path and a measure to remove paths of classes with no instance path are proposed.
Related application includes Visor[
We have been developing a
practical LOD search tool named
SPARQL Builder for the
lifescience data analysis (http://
www.sparqlbuilder.org/). This
tool provides an interactive GUI
that allows users who are not
familiar with SPARQL language to
generate SPARQL queries without
knowledge of SPARQL and RDF
data schema [
SPARQL Builder manages SBM generated by accessing SPARQL endpoints in advance (1). When Fig. 1. Overview of the SPARQL Builder sysa user access to the SPARQL tem.
Builder system via a web browser as a GUI, SPARQL Builder obtains a list of classes by analysing SBM (2) and displays the list on the user's web browser (3). Then, when the user selects "input" and "output" classes, SPARQL Builder constructs class paths by traversing the class graph constructed using information described in SBM (4) and draw them on the web browser. Using this GUI, users can explore datasets as their interest by specifying classes. If a user interested in the interrelationships between molecular pathways and proteins, he should do at rst is to select Protein as input class and Pathway as output class. Then, SPARQL Builder shows all possible paths involves pathways in which proteins that catalyses chemical reactions constitutes. These paths has sequentially connected two relationships as the form of "Protein -(left/right)BiochemicalReaction -(pathwayComponnt)- Pathway". When he select one of the class paths, SPARQL Builder create a SPARQL query which can use to retrieve data his interest. SPARQL Builder is used for support service to generate SPARQL queries for 38 SPARQL endpoints as of July 2016. 3
SPARQL Builder Metadata (SBM), is a summary of RDF datasets provided via a SPARQL endpoint. SBM is de ned as an extension of VoID (https://www.w3. org/TR/void/) and SPARQL 1.1 service description (https://www.w3.org/TR/ sparql11-service-description/) with our original vocabulary whose name space is sbm:. SBM contains statistic summary data called \graph summary" for default graph and each named graph provided by the SPARQL endpoint. Graph summary is an extension of VoID vocabulary related to void:Dataset class with detailed statistical parameters as follows: A property partition is a subset of RDF dataset associated with a property. In addition to original VoID properties, three properties sbm:subjectClasses, sbm:objectClasses, and sbm: objectDatatypes to describe numbers of classes and datatypes are used. A class relation is a distinct pair of a subject class and an object class/datatype, where subject class and object class/datatype are the class of subject instances and class/datatype of object instances/literals in all triples associated with the concerned property partition. sbm:classRelation property is introduced to describe each class relation with properties sbm:subjectClass, sbm:objectClass, and sbm:objectDatatype as our original extension and properties VoID vocabulary. 4
To compute paths between two classes e ciently, we employed a specialized
graph whose nodes and edges correspond to classes and the class{class relations
with predicates, respectively. We call the graph class graph. A class graph can be
constructed from SBM e ciently because SBM includes a list of all the classes
and a list of all the class{class relationships. Given a class graph, an undirected
path on the graph is called as a class path. Note that a class path is not always
simple path because the same classes may appear twice or more in the path with
di erent properties. Class paths between two classes can be found in practically
short time using algorithm written in [
We introduced SPARQL Builder which enables practical LOD data searching in a SPARQL endpoint. Although the system originally was designed for biological databases, the technologies used in the system including SBM and class graphs are applicable to another domain. Therefore, our future work includes expanding our application into multiple domains and evaluate the generalities of our approach. In addition, we will consider to expand class paths into more general types of subgraphs on class graph, to support more styles of SPARQL queries. In addition, supporting federated search also remains as future work. Acknowledgments This work was supported by JSPS KAKENHI Grant Number 25280081, 24120002 and the National Bioscience Database Center (NBDC) of the Japan Science and Technology Agency (JST).