RDF2TG: Towards Supporting RDF in TigerGraph Property Graph Database System

Both RDF and property graph models provide ways to explore and graphically depict connected data. But the two graph models are very different, while each has different strengths in different use cases. RDF is a standard model for data interchange on the Web. RDF has features that facilitate data integration even if the underlying schemas of data sources are different. With the support of RDF Schema (RDFS) and Web Ontology Language (OWL), it enables the inference engine to discover more implicit knowledge in the graph. In contrast, property graphs were developed about efficient storage that would allow for fast querying and traversals across connected data. However, once after choosing one graph model, it is not convenient to losslessly transfer it to another one due to different reasons, e.g, time and cost consumption, technical challenges in converting between two graph models, and different query languages. To provide solutions to customers wanting to migrate their existing RDF graph models to property graph models, we propose a method to migrate LBDC Semantic Publishing RDF Benchmark (SPB) to property graphs using TigerGraph. In addition to mapping about 32 million triples, we also translate 36 SPARQL queries to GSQL, which is the query language used in the TigerGraph database, with an initial analysis of query performance.

Reproducibility

• LDBC SPB benchmark generator instruction

  Please see details in SPB sepecification We upload the generated RDF graph in turtle format which can be directly used.

• Create Schema, Map data, Load CSV file, Run queries, Evaluate performance

  Run the ipynb file in Jupyter Notebook or run the python code directly

note: at least 30GB memory is required to run the project

Data Sources

• original ldbc website: LDBC Semantic Publishing Benchmark
• data sets used in this project: rdf dump in turtle and data files in csv

Dependencies

rdflib, pytigergraph, pandas, csv

Graph Model Mapping

Mapping Rules and examples from RDF graphs to Tigergraph

@prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#> 
@prefix ex: <http://example.com/> 
@prefix xsd: <http://www.w3.org/2001/XMLSchema#>

ex:Tigergraph rdf:type ex:Private Company .
ex:Tigergraph ex:hasName "Tigergraph"^^xsd:string .
ex:Tigergraph ex:hasName "Tigergraph"@en .

Mapping Rules	RDF Graphs	TigerGraph	Examples
Rule 1	Subject	ClassInstance	ex:Tigergraph
Rule 2	Predicate (Object Property)	ObjectPropertyInstance	rdf:type
Rule 3	Predicate (Datatype Property)	DatatypePropertyInstance	ex:hasName
Rule 4	Object (!isLiteral)	ClassInstance	ex:Private Company
Rule 5	Object (isLiteral)	ValueInstance	"Tigergraph"^^xsd:string
Rule 6	Literal Value	ValueInstance's attribute: value	Tigergraph
Rule 7	Datatype	ValueInstance's attribute: datatype	xsd:string
Rule 8	Language Tag	ValueInstance's attribute: langTag	en

Example Graphs in Diagram

Figure 1. Example triples in RDF Graph

Figure 2. Example triples in Tigergraph

SPARQL to GSQL Translation

Mappings and Examples:

Mapping Rules	SPAQRL	GSQL	SPAQRL Examples	GSQL Examples
Rule 1	SELECT	SELECT	SELECT ?cwork	SELECT s.uri as cwork INTO T
Rule 2	WHERE	FROM & WHERE	WHERE {?cwork cwork:dateModified ?dateModified. }	FROM ClassInstance:s - (hasDatatypePropertyInstance>:e) - DatatypePropertyInstance:t - (hasValueInstance>:e1) - ValueInstance:t1 WHERE t.uri == "cwork:dateModified"
Rule 3	FILTER	WHERE	FILTER (?dateModified >= "2011-09-01T00:00:00.000"^^xsd:dateTime)	WHERE to_datetime(t1.value) >= to_datetime("2011-09-01T00:00:00.000")
Rule 4	OPTIONAL	ACCUM CASE	OPTIONAL {?cwork cwork:title ?title .}	ACCUM CASE t.uri WHEN "cwork:title" THEN s.@title += t1.value
Rule 5	GROUP BY	GROUP BY	GROUP BY ?cwork	GROUP BY s.uri
Rule 6	ORDER BY	ORDER BY	ORDER BY ASC(?dataModified) ASC(?cwork)	ORDER BY s.@dateModified ASC, cwork ASC
Rule 7	LIMIT	LIMIT	LIMIT 5	LIMIT 5

Example query description:

Retrieve properties dateModified, title, category, liveCoverage, audience for all 
creative works that are of a given type. The value of property dateModified of the 
retrieved creative works should be within a certain time range. Return 5 results 
ordered in ascending order by their dateModified.

Figure 3. SPARQL query

Figure 4. Results returned by running SPARQL query

Figure 5. GSQL query

Figure 6. Results returned by running GSQL query

Query Performance Evaluation

• Platform:

Configuration	Value
Graph DBMS	TigerGraph & RDF triple store (anonymous)
OS	MacOS 12.4 Montery
CPU	2.3GHz 8-core Intel i9
Memory	30G 2667 MHz DDR4

• Queries:

  11 basic and 25 advanced SPARQL and GSQL queries

• Matrix:

  Average of Execution Time (10 times)

• Result: Average of Execution Time in TG/Average of Execution Time in RDF triple store (no engine level optimization)

  • 44.4% Faster in Tigergraph (1.5 - 83.9 SpeedUp)
  • 27.8% Faster in RDF triplestore (1.2 - 8.7 SpeedUp)
  • 27.8% very close (within 0.03s)

Contributers

Lu Zhou, Jay Yu 
Innovation and Development Center
TigerGraph, Inc.