Transforming RDF-star to Property Graphs A Preliminary Analysis of Transformation Approaches extended version

2025-05-01 1 0 1.39MB 40 页 10玖币

侵权投诉

Transforming RDF-star to Property Graphs:

A Preliminary Analysis of Transformation

Approaches – extended version

Ghadeer Abuoda1, Daniele Dell’Aglio1, Arthur Keen2, and Katja Hose1

1Department of Computer Science, Aalborg University, Aalborg, Denmark

{gsmas,dade,khose}@cs.aau.dk

2ArangoDB, San Francisco, United States

arthur@arangodb.com

Abstract. RDF and property graph models have many similarities,

such as using basic graph concepts like nodes and edges. However, such

models diﬀer in their modeling approach, expressivity, serialization, and

the nature of applications. RDF is the de-facto standard model for knowl-

edge graphs on the Semantic Web and supported by a rich ecosystem

for inference and processing. The property graph model, in contrast,

provides advantages in scalable graph analytical tasks, such as graph

matching, path analysis, and graph traversal. RDF-star extends RDF

and allows capturing metadata as a ﬁrst-class citizen. To tap on the ad-

vantages of alternative models, the literature proposes diﬀerent ways of

transforming knowledge graphs between property graphs and RDF. How-

ever, most of these approaches cannot provide complete transformations

for RDF-star graphs. Hence, this paper provides a step towards trans-

forming RDF-star graphs into property graphs. In particular, we identify

diﬀerent cases to evaluate transformation approaches from RDF-star to

property graphs. Speciﬁcally, we categorize two classes of transforma-

tion approaches and analyze them based on the test cases. The obtained

insights will form the foundation for building complete transformation

approaches in the future.

1 Introduction

The most popular models for representing knowledge graphs are: RDF1(Re-

source Description Framework) and property graphs [27] (PG). While RDF rep-

resents knowledge graphs as a set of subject-predicate-object triples, property

graphs assign key-value style properties to nodes and edges. Recently, RDF-

star [18] has been proposed as an extension of RDF to enable enriching RDF

triples with metadata information by embedding triples in subjects or objects of

other triples, which allows providing statements about statements and somewhat

resembles adding properties to edges in property graphs. RDF-star is supported

by a rich ecosystem of data management systems and standards, most notably

1RDF 1.1 Primer: https://www.w3.org/TR/rdf11-primer/

arXiv:2210.05781v1 [cs.DB] 11 Oct 2022

systems such as Stardog, OpenLink’s Virtuoso, Ontotext GraphDB, Allegro-

Graph, Apache Jena, and more recently also Oxigraph, but also query standards,

such as SPARQL2and its extension SPARQL-star3as well as RDF Schema,

which allows describing classes of RDF resources and properties4. In contrast,

many graph database systems, such as Neo4j, TigerGraph, JanusGraph, Redis-

Graph, and SAP HANA are based on diﬀerent variations of the property graph

model [31] and diﬀerent query languages [9,10]. Unfortunately, RDF-star graphs

and property graphs are not entirely compatible with one another. Although they

both describe data through graphs, their underlying models and semantics are

diﬀerent, leading to many data interoperability issues [3, 22]. Metadata or edge

properties in RDF-star can be modeled as separate nodes or RDF-star triples.

In contrast, edge properties can only be represented as literal key-value pairs in

property graphs. In general, it is challenging to transform an RDF-star graph

fully into a property graph because of the rich expressiveness of the former. The

heterogeneity between the two models and their frameworks makes it necessary

to study their interoperability, i.e., the ability to map one model to another for

data exchange and sharing [3].

The mapping between the two models is crucial for data exchange, data in-

tegration as well as reusability of systems and tools between the frameworks.

RDF-star, speciﬁcally the RDF model, is recognized as a web-native model that

supports data exchange and sharing across diﬀerent sources because of its formal

semantics and the universal uniqueness of resources using IRIs. RDF is a com-

mon and ﬂexible model for knowledge representation, and that is exempliﬁed

by knowledge graphs that cover a broad set of domains, such as DBpedia [23],

YAGO [30], and Wikidata [33]. On the contrary, even with the wide adoption

of property graph engines, property graphs lack many essential features, such as

a schema language, a standard query language, standard data serialization for-

mats, etc. Achieving interoperability and reliable transformations between the

two frameworks will ﬁnally enable us to exploit the beneﬁts of both models.

The transformation of property graphs to RDF-star has been explored re-

cently [21], and basic transformation rules for property graphs to RDF-star were

proposed [4,18]. However, the latter does not cover all RDF-star constructs and

allows for multiple alternatives.

Listing 1.1: An example RDF-star graph in Turtle star format

@prefix ex: < htt p :/ / e xa mpl e . or g /> .

<< ex:Alex ex:a ge 25 > > ex: c e rtai nty 0.5 .

Consider, for instance, the example illustrated in Figure 1(a). If we start

with the triple (ex:Alex,ex:age,25), then we could represent the RDF ele-

ment (Alex) as a node in a property graph, as shown in Figure 1(b). This node

would then have a property (age,25) and the RDF triple would be represented

2SPARQL 1.1 Query Language: https://www.w3.org/TR/sparql11-query/

3SPARQL-star Query Language: https://w3c.github.io/rdf-star/cg-spec/editors_draft.html#

sparql-star

4RDF Schema 1.1: https://www.w3.org/TR/rdf-schema/

:California

Alex

:Apple_Inc

:CEO

certainty

:start_date

0.5

age

Alex

(age = 25)

(a) RDF-star Graph

:California

Alex

:Apple_Inc

:CEO

certainty

:start_date

0.5

age

Alex

(age = 25)

(b) A property graph

Fig. 1: Graphical Representation of Listing 1.1 as (a) RDF-star and (b) Property

Graph

by a single node in a property graph. However, if we have a single node with-

out an edge, we cannot represent the metadata about the original RDF triple

(ex:certainty 0.5) in the property graph. Studying such cases, this paper makes

the following contributions:

–We identify two alternative approaches of transformations: RDF-topology-

preserving and Property-Graph transformation.

–We deﬁne a set of test cases capturing the diverse RDF-star constructs that

have to be considered when transforming RDF-star to property graphs.

–Using the test cases, we systematically evaluate alternative mapping ap-

proaches and identify their shortcomings.

This paper is structured as follows: while Section 2 introduces preliminaries,

Section 3 discusses related work. Section 4 presents alternative transformation

approaches. Afterwards, Section 5 provides details on our test cases, which we use

in Section 6 to identify and discuss shortcomings of transformation approaches.

Section 7 concludes the work with an outlook for future work.

2 Preliminaries

In this section, we formally introduce RDF1, RDF-star [18], and property

graphs [27].

Resource Description Framework (RDF). RDF is a W3C standard data

model that represents information as a set of statements. Each statement denotes

a typed relation between two resources.

Deﬁnition 1 (RDF statement). Let I,Band Lbe the disjoint sets of In-

ternationalized Resource Identiﬁers (IRIs), blank nodes and literals. An RDF

statement is a triple (s, p, o)∈(I∪B)×I×(I∪B∪L), and it indicates that s

and o(subject and object, resp.) are in a relation p(predicate).

In this paper, we consider two types of RDF statements that we distinguish

based on whether the object is an IRI or a literal. Object property statements are

RDF statements (s, p, o)∈(I∪B)×I×(I∪B), while datatype property statements

:California

:Tim_Cook

:Apple_Inc

:CEO

:located_in

:start_date

2011

(a) RDF Graph

:California

:Tim_Cook

:Apple_Inc

:CEO

:located_in

:start_date

2011

(b) RDF-star Graph

:California

:Tim_Cook

:Apple_Inc :CEO

:located_in

(start_date= 2011)

Fig. 2: Example in RDF, RDF-star, and Property Graphs

are RDF statements (s, p, o)∈(I∪B)×I×L. An RDF graph containing three

RDF statements is shown in Listing 1.2 serialized in Turtle5, and visually in

Figure 2(a) as a graph. The ﬁrst two statements are object property statements.

The ﬁrst statement describes two resources, ex:Apple_Inc and ex:California,

related by the predicate ex:located_in. The second statements indicates that

ex:Apple_Inc has ex:Tim_Cook as a ex:CEO. The last statement is a datatype

property statement, and it indicates that ex:Tim_Cook has the literal "2011" as

the value of the ex:start_date predicate.

Listing 1.2: An RDF graph in Turtle format

@prefix ex: < htt p :/ / e xa mpl e . or g /> .

ex: Apple_Inc ex: l ocat ed_i n ex: Cali forn i a .

ex: Apple_Inc ex:CEO ex:Ti m_C ook .

ex: T i m_C ook ex:s t art _ date 2011 .

RDF-star Looking at the RDF graph in Figure 2(a), one can spot some im-

precise data modelling choices: stating that Tim Cook started in 2011 is not

totally correct, as he started in 2011 his role as CEO of Apple. In other words,

one should associate the starting date to the statement (ex:Apple_Inc,ex:CEO,

ex:Tim_Cook), as depicted in Figure 2(b). There are several ways to imple-

ment this idea in RDF, such as RDF reiﬁcation [18], singleton properties [25],

and named graphs [7]. However, these mechanisms have signiﬁcant shortcom-

ings [11, 16, 19, 26].

Listing 1.3: An RDF-Star Graph in Turtle-Star Format

@prefix ex: < htt p :/ / e xa mpl e . or g /> .

ex: Apple_Inc ex: l ocat ed_i n ex: Cali forn i a .

<< ex: Apple_Inc ex: CEO ex: Tim_Cook >> ex: s t art_ d ate 2011 .

A solution to overcome such shortcomings was recently proposed by Hartig et

al. with RDF-star [16,18]. RDF-star extends RDF by letting RDF statements be

subjects or objects in other statements. Listing 1.3 shows an RDF-star document

serialized in Turtle-star [18]. The ﬁrst statement is a compliant RDF statement

5RDF Turtle: https://www.w3.org/TR/turtle/

(it appeared also in Listing 1.2). The second statement indicates that ex:Apple_

Inc appointed ex:Tim_Cook as ex:CEO in 2011. We formally deﬁne an RDF-star

statement as follows.

Deﬁnition 2 (RDF-star statement). Let s∈I∪B,p∈I,o∈I∪B∪L.

An RDF-star statement is a triple deﬁned recursively as:

–Any RDF statement (s, p, o)is an RDF-star statement;

–Let tand ¯

tbe RDF-star statements. Then, (t, p, o),(s, p, t)and (t, p, ¯

t)are

RDF-star statements, also known as asserted statement.tand ¯

tare called

embedded or quoted statements.

Property Graphs A property graph (PG) is a graph where nodes and edges can

have multiple properties, represented as key-value pairs. Figure 2(c) illustrates

the graph described in the above section as a property graph. In this case, the

starting date of Tim Cook as the CEO of Apple Inc is reported as a key-value

property on the :CEO edge. PGs have not a unique and standardized model; each

PG engine proposes its data model. A generic PG model deﬁnition is proposed

by [31].

Deﬁnition 3 (Property Graph). Let Lbe the set of the labels, P N be the set

of property names, and Dbe the set of property values. A property graph Gis

an edge-labeled directed multi-graph such that G= (N, E, edge, lbl, P, σ), where:

–Nis a set of nodes,

–Eis a set of edges between nodes, such that N∩E=∅

–edge :E→(N×N)is a total function that associates each edge in Ewith

a pair of nodes in N. If edge(e1)=(n1, n2),n1is the source node and n2is

the target node.

–lbl : (N∪E)→ P(L)is a function that associates each edge or node with a

set of labels.

–σ: (N∪E)→ P(P)is a function that associates a node or edge with a

non-empty set of properties Pdeﬁned as a set of key-value pairs (k, v) where

k∈P N and v∈D

To ease all approaches’ output representation, we map any IRI to a distinct

string representing a local name. Given I, a set of all IRIs, localN ame is a

function that maps an IRI to a string that represents the local name of an RDF

resource6. For example, the local name for the RDF resource (http://example.

com/meets)localName("http://example.com/meets") is "meets". We will use

this function in the output representation in Section 6.

3 Related Work

We can distinguish between related work on converting between (i) RDF and

PG and (ii) RDF-star and PG.

6In Neo4j, the user can conﬁgure the local name of RDF terms such as subProper-

tyOf, subClassOf, Class, etc.

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

TransformingRDF-startoPropertyGraphs:APreliminaryAnalysisofTransformationApproachesextendedversionGhadeerAbuoda1,DanieleDell'Aglio1,ArthurKeen2,andKatjaHose11DepartmentofComputerScience,AalborgUniversity,Aalborg,Denmark{gsmas,dade,khose}@cs.aau.dk2ArangoDB,SanFrancisco,UnitedStatesarthur@arangodb.c...

展开>> 收起<<

Transforming RDF-star to Property Graphs A Preliminary Analysis of Transformation Approaches extended version.pdf

共40页,预览5页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Transforming RDF-star to Property Graphs A Preliminary Analysis of Transformation Approaches extended version

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: