=Paper=
{{Paper
|id=Vol-2293/jist2018pd_paper8
|storemode=property
|title=Mapping RDF Graphs to Property Graphs
|pdfUrl=https://ceur-ws.org/Vol-2293/jist2018pd_paper8.pdf
|volume=Vol-2293
|authors=Shota Matsumoto,Ryota Yamanaka,Hirokazu Chiba
|dblpUrl=https://dblp.org/rec/conf/jist/MatsumotoYC18
}}
==Mapping RDF Graphs to Property Graphs==
https://ceur-ws.org/Vol-2293/jist2018pd_paper8.pdf
Mapping RDF Graphs to Property Graphs
Shota Matsumoto1 , Ryota Yamanaka2 , and Hirokazu Chiba3
1
Lifematics Inc., Tokyo 101-0041, Japan
shota.matsumoto@lifematics.co.jp
2
Oracle Corporation, Bangkok 10500, Thailand
ryota.yamanaka@oracle.com
3
Database Center for Life Science, Chiba 277-0871, Japan
chiba@dbcls.rois.ac.jp
Abstract. Increasing amounts of scientific and social data are published
in the Resource Description Framework (RDF). Although the RDF data
can be queried using the SPARQL language, even the SPARQL-based
operation has a limitation in implementing traversal or analytical algo-
rithms. Recently, a variety of graph database implementations dedicated
to analyses on the property graph model have emerged. However, the
RDF model and the property graph model are not interoperable. Here,
we developed a framework based on the Graph to Graph Mapping Lan-
guage (G2GML) for mapping RDF graphs to property graphs to make
the most of accumulated RDF data. Using this framework, graph data
described in the RDF model can be converted to the property graph
model and can be loaded to several graph database engines for further
analysis. Future works include implementing and utilizing graph algo-
rithms to make the most of the accumulated data in various analytical
engines.
Keywords: RDF · Property Graph · Graph Database
1 Introduction
Increasing amounts of scientific and social data are described as graphs. As a
format of graph data, the Resource Description Framework (RDF) is widely used.
Although RDF data can be queried using the SPARQL language in a flexible
way, SPARQL is not dedicated to traversal of graphs and has a limitation in
implementing graph analysis algorithms.
In the context of graph analysis, the property graph model [1] is becom-
ing popular; various graph database engines, including Neo4j [2], Oracle Labs
PGX [3], and Amazon Neptune [4], adopt this model. These graph database en-
gines support algorithms for traversal or analyzing graphs. However, currently
not many datasets are consistently described in the property graph model, so
the application of these powerful engines are limited.
Considering this situation, it is valuable to develop a method to transform
RDF data into property graphs. However, the transformation is not straightfor-
ward due to the differences in the data model. In RDF graphs, all information is
2 S. Matsumoto et al.
expressed as the triple (node-edge-node), whereas in property graphs, arbitrary
information can be contained in each of the nodes and edges as key-value form.
Although previous works addressed this issue by formalizing transformations [5],
users cannot define their specific mappings intended for each use case.
Here, we developed a framework based on the Graph to Graph Mapping
Language (G2GML) for mapping RDF graphs to property graphs. Using this
framework, accumulated graph data described in the RDF model can be con-
verted to the property graph model and can be loaded to several graph database
engines.
2 Methods
Figure 1 shows the overview of proposed framework. In the proposed framework,
users write mappings from RDF graphs to property graphs in G2GML. This
mapping can be processed by an implementation called G2G Mapper, which is
implemented by authors (available on https://github.com/g2gml). This tool
retrieves RDF data from SPARQL endpoints and converts them to property
graph data in several different formats specified by popular graph databases.
G2GML is a declarative language which consists of pairs of RDF graph pat-
terns and property graph patterns. An intuitive meaning of a G2GML is a map-
ping between RDF subgraphs that matches the described patterns and described
components of the property graph. In the next section, we briefly explain the
syntax of G2GML with a concrete example usage.
Fig. 1. Overview of G2GML mapping
3 Example
Figure 2 shows an example of G2GML mapping, which converts RDF data re-
trieved from DBpedia into property graph data. When we focus on relationships
Mapping RDF Graphs to Property Graphs 3
Fig. 2. Mapping of RDF data
that one musician and another are in the same group, the information can be
summarized into the property graph data as shown in this figure.
For this conversion, the actual G2GML is described as in Figure 3. It starts
with URI prefixes used to write mappings, and then, each mapping consists of
one unindented line of a property graph pattern and indented lines of an RDF
graph pattern. A property graph pattern is written in a syntax like Cypher
(the query language of Neo4j), whereas an RDF graph pattern is written as a
pattern in SPARQL. Variables in each pattern are mapped by those names. This
example contains one node mapping for Musician entity and one edge mapping
for same group relationship only. In G2GML, edge mappings are defined based
on the conditions of node mappings, which means that edges are generated
in property graph iff both nodes’ patterns and edges’ patterns are matched in
RDF graph. Also, mus, nam, dat, twn and len are used as variables to extract
resources and literals from RDF graph. In the resulting property graph, resources
can be mapped to nodes, while literals can be mapped to values of properties.
Finally, Figure 4 shows the SPARQL query to retrieve the pairs of musicians
who are in the same group. After G2GML mapping above, we can load the
generated property graph data into graph databases, such as Oracle Labs PGX,
and the query can be written in PGQL (the query language of PGX).
4 Conclusion
In this work, we defined G2GML for mapping RDF graphs to property graphs
and implemented a converter based on the G2GML. We also showed an example
usage of G2GML. Future works include further analysis of the converted graph
data on the database engines adopting the property graph model.
4 S. Matsumoto et al.
# Prefixes
PREFIX rdf:
PREFIX rdfs:
PREFIX prop:
PREFIX schema:
PREFIX dbpedia-owl:
PREFIX foaf:
# Node mapping
(mus:Musician {vis_label:nam, born:dat, hometown:twn}) # PG Pattern
?mus rdf:type foaf:Person, dbpedia-owl:MusicalArtist . # RDF Pattern
?mus rdfs:label ?nam .
OPTIONAL { ?mus prop:born ?dat }
OPTIONAL { ?mus dbpedia-owl:hometown / rdfs:label ?twn }
# Edge mapping
(mus1:Musician)-[:same_group {label:nam, length:len}]->(mus2:Musician) # PG Pattern
?grp a schema:MusicGroup ; # RDF Pattern
dbpedia-owl:bandMember ?mus1 , ?mus2 .
FILTER(?mus1 != ?mus2)
OPTIONAL { ?grp rdfs:label ?nam. FILTER(lang(?nam) = "ja")}
OPTIONAL { ?grp dbpedia-owl:wikiPageLength ?len }
Fig. 3. G2GML mapping definition
# SPARQL
PREFIX rdf:
PREFIX rdfs:
PREFIX schema:
PREFIX dbpedia-owl:
SELECT DISTINCT
?nam1 ?nam2
WHERE {
?mus1 rdf:type foaf:Person , dbpedia-owl:MusicalArtist .
?mus2 rdf:type foaf:Person , dbpedia-owl:MusicalArtist .
?mus1 rdfs:label ?nam1 . FILTER(lang(?nam1) = "ja") .
?mus1 rdfs:label ?nam2 . FILTER(lang(?nam2) = "ja") .
?grp a schema:MusicGroup ;
dbpedia-owl:bandMember ?mus1 , ?mus2 .
FILTER(?mus1 != ?mus2)
}
# PGQL
SELECT DISTINCT m1.name, m2.name WHERE (m1)-[same_group]-(m2)
Fig. 4. SPARQL and PGQL
References
1. Angles, R., Arenas, M., Barcel, P., Hogan, A., Reutter, J., Vrgoc, D.: Foundations of
Modern Query Languages for Graph Databases. ACM Computing Surveys (CSUR),
50(5), 68 (2017)
2. The Neo4j Graph Platform, https://neo4j.com/.
3. Oracle Labs Parallel Graph AnalytiX (PGX), https://www.oracle.com/
technetwork/oracle-labs/parallel-graph-analytix/overview/index.html
4. Amazon Neptune, https://aws.amazon.com/neptune/.
5. Hartig, O.: Reconciliation of RDF* and property graphs. arXiv preprint
arXiv:1409.3288 (2014)