Metadata publication in accordance with the semantic web as a database is a trend for providing and integrating various life-science data. These metadata are published as SPARQL endpoints, a standardised API for RDF datasets. As life-science data are very widely diverse and described using various ontologies and data classes, writing an e cient SPARQL query for SPARQL endpoints is di cult for biologists. To address this problem, we propose an intelligent SPARQL query builder that enables users to build a query without knowledge of SPARQL or the data schema. We have developed a prototype version of the SPARQL builder accessible via users' web browsers. The system crawls SPARQL endpoints in advance to analyse the data schema of large amounts of data, and the resultant crawled data are stored as RDF datasets. This paper focuses on the implementation including the system overview, and the data structure of the resultant crawled data.
With the development of life-science research elds and measurement
technologies for biological phenomena, the diversity of research data has been
increasing. For e cient circulation, intelligent analysis and integration of such
heterogeneous data, semantic web technologies including RDF and SPARQL have
been adapted, and life-science metadata datasets have already been published as
SPARQL endpoints such as the European Bioinformatics Institute (EBI) RDF
platform [
To address this problem, we propose an intelligent web tool named SPARQL Builder that enables users to build a SPARQL query without understanding RDF data schema or SPARQL. We have implemented a prototype version of SPARQL Builder that enables users to build a SPARQL query for existing lifescience SPARQL endpoints, including EBI's service. This paper reports the implementation issues of the prototype system. 2
SPARQL Builder is an intelligent tool that assists a user with no knowledge of SPARQL to generate a query on the basis of a triple path. To be more precise, a triple path i1 p!1 i2 p!2 : : : p!n in+1; (1 n, and n = 3 is our default) is a sequence of instances i1; i2; : : : in+1 of classes C1; C2; : : : ; Cn+1 respectively, connected by properties p1; p2; : : : pn. A list C1; C2; : : : ; Cn+1 of classes is called a class path if a triple path for the list exists. When a SPARQL endpoint, a start class C1 and an end class Cn+1 are speci ed by a user on the system, the system analyses the metadata of the SPARQL endpoint obtained by a crawler in advance (cf. Section 3) and displays possible class paths C1; C2; : : : ; Cn+1. A user further selects a class path. Then the system generates a SPARQL query that searches a triple path corresponding to the selected class path.
Figure 1 shows part of a screen capture of the SPARQL Builder client performing on a web browser. Our prototype system is implemented as a Java servlet, and a user can access through its client written in JavaScript using a web browser. To obtain possible class paths between the user's start and end classes in a practical time, we use a data schema for the SPARQL endpoints called endpoint metadata. To construct the endpoint metadata for a SPARQL endpoint, SPARQL Builder throws small but numerous SPARQL queries to the endpoint in advance. The endpoint metadata are written in the vocabulary called SPARQL Builder metadata, published at http://sparqlbuilder.org/doc/, and they are stored in the servlet server in RDF. As of September 2014, we have retrieved endpoint metadata from EBI's ve SPARQL endpoints of large-scale databases used by the most cutting-edge research, including Expression Atlas1, BioModels2, BioSamples3, ChEMBL4 and Reactome5.
Though our SPARQL Builder itself is an individual application, it is designed
to work in conjunction with TogoTable [
SPARQL Builder metadata brie y and comprehensively describes an RDF graph schema of SPARQL endpoint datasets. Other speci cations de ned for a similar purpose include the vocabulary of interlinked datasets VoID 6 and the vocabulary for describing SPARQL services SPARQL 1.1 Service Description7. SPARQL Builder metadata is based on these existing speci cations but is dened by adding our original vocabularies that describe metadata for constructing class paths and statistics to determine comprehensiveness of the data that can
be handled by our search method on the basis of class paths. For instance, our original class ClassRelation is used to describe a relationship of two classes correlated with property p that is essential to build a class path. In order to improve comprehensiveness of our triple path search, the class{class relationship as a ClassRelation is not only the domain and range classes of property p but also the classes of subject and object instances of triples having property p.
As described above, SPARQL Builder metadata is our original speci cation, but it is de ned for arbitrary SPARQL endpoints. Some life-science SPARQL endpoints provide metadata for their datasets. EBI publishes such metadata in their framework called Lodestar8, and Bio2RDF publishes Bio2RDF Dataset Metrics9. We hope these metadata speci cations are integrated as a global standard and promote distribution of metadata for advanced intelligent semantic web data processing. 4
We discussed our prototype version of the SPARQL Builder tool, which enables users to discover a sequentially connected triple path for arbitrary SPARQL endpoint without knowledge of the data schema or SPARQL. In order to a build SPARQL query by interaction with a user in a practical time, the metadata of the datasets provided by SPARQL endpoints are retrieved in advance and the results are stored as RDF datasets followed by a SPARQL Builder metadata speci cation. Our future work includes support for SPARQL queries not only for triple paths of a sequence of instances but also general structures such as trees, veri cation and improvement of practicality of our prototype system. Acknowledgments. We thank Dr Yasunori Yamamoto for useful comments for improvement of the SPARQL Builder metadata speci cation.