Exploring RDF Datasets with LDscout

Anna Goy, Diego Magro and Francesco Conforti

Dipartimento di Informatica, Universit

a di Torino, C. Svizzera 185, Torino, Italy

Keywords:

Semantic Web, Linked Data, RDF Dataset, Dataset Exploration.

Abstract:

In this paper, we present the formal model underlying LDscout, a Java library enabling developers to query a

dataset specifying the vocabulary (ontology) they want to use and the instances they want to query about. The

model is based on the formal deﬁnition of the concepts of Exploration Task and Exploration Task Solution,

and is independent from the dataset. In this paper, we present the speciﬁc implementation that enables the

interaction with RDF triplestores using OWL ontologies. In order to assess our approach, we report the

usage of LDscout within PRiSMHA, a Digital Humanities project aimed at enhancing the access to historical

archives through Semantic Web technologies.

1 INTRODUCTION

The Linked Open Data (LOD) paradigm (Heath and

Bizer, 2011) has become a proﬁtable approach in

many ﬁelds, in a twofold perspective: (a) for pu-

blishing data, to gain visibility for the large amount

of information otherwise buried in proprietary repo-

sitories; (b) for using data, to leverage the informa-

tion widely available in a structured and machine-

readable way. The number of available datasets in the

LOD cloud is constantly increasing (see https://lod-

cloud.net/). Among the ﬁelds that took up LOD

practices, ranging from e-government to healthcare,

cultural heritage has started playing a major role

(Alexiev, 2016) (Edelstein et al., 2013); see also the

ArCo project in Italy: http://dati.beniculturali.it.

According to the Linked Data paradigm, data are

expressed as RDF triples (W3C, 2014) and RDF data-

sets are accessed through SPARQL endpoints (W3C,

2013). This structure implies that, in order to query

LOD datasets and to ”understand” the query results,

clients – being either users or applications – need to

be familiar with the query language, i.e. SPARQL,

and to understand the vocabulary used in the datasets.

In Section 2 we will provide a brief survey of

the approaches providing friendly User Interfaces ai-

med at supporting human users in overcoming the

complexity of SPARQL queries and the difﬁculties in

grasping the semantic model underlying RDF data-

sets. In this paper, instead, we focus on the provision

of suitable APIs, aimed at offering software clients

(applications) an easy and ﬂexible way to access LOD

datasets. In particular, we will present the formal mo-

del underlying LDscout, a Java library enabling deve-

lopers to specify the vocabulary they can understand

(or they are currently focusing on) and the instances

they want to query about. The library is wrapped in

a RESTful web service interface, in order to offer a

language-independent interoperability.

LDscout is based on a general model, that can be

used to query any dataset. The speciﬁc implementa-

tion we present here enables the interaction with RDF

datasets and in order to assess its usefulness, we used

it within PRiSMHA (Providing Rich Semantic Me-

tadata for Historical Archives: di.unito.it/prismha).

PRiSMHA is a Digital Humanities project aimed at

experimenting with a crowdsourcing approach for the

construction of ontology-based formal semantic re-

presentations of the content of historical documents

(Goy et al., 2017), within the overall goal of enhan-

cing the access to historical archives through Seman-

tic Web technologies (Goy et al., 2015). The project is

funded by Compagnia di San Paolo and Universit

a di

Torino and relies on a collaboration between Compu-

ter Science and Historical Studies departments of the

same University with Fondazione Istituto Piemontese

A. Gramsci, member of the Polo del ’900 foundation

(www.polodel900.it). The current proof-of-concept

prototype we are developing within the project in the

area of cultural heritage deals with archival resour-

ces concerning the students and workers protest du-

ring the years 1968-1969 in Italy.

The paper is organized as follows. In Section 2 we

brieﬂy survey the relevant work investigating the inte-

Goy, A., Magro, D. and Conforti, F.

Exploring RDF Datasets with LDscout.

DOI: 10.5220/0006957600920100

In Proceedings of the 10th International Joint Conference on Knowledge Discovery, Knowledge Engineer ing and Knowledge Management (IC3K 2018) - Volume 3: KMIS, pages 92-100

ISBN: 978-989-758-330-8

raction with RDF datasets. Section 3 presents the mo-

del underlying LDscout by providing a general formal

deﬁnition of Exploration Task (Section 3.1), its in-

stantiation on RDF triplestores and OWL ontologies

(Section 3.2), and the mechanisms for computing so-

lutions to Exploration Tasks (Section 3.3); the section

is concluded by summarizing how we used LDscout

within the platform developed in the PRiSMHA pro-

ject (Section 3.4). Finally, Section 4 mentions the

main directions of our future work.

2 RELATED WORK

”The term Linked Data refers to a set of best practices

for publishing and interlinking structured data on the

Web” (Ngonga-Ngomo et al., 2014), p.1. The intro-

duction to Linked Data by Ngonga Ngomo and collea-

gues contains a detailed overview of the Linked Data

life-cycle and discusses the state-of-the-art of diffe-

rent aspects of it (namely, extraction, authoring, lin-

king, enrichment, and quality assessment). As alre-

ady mentioned in Section 1, in order to query RDF

datasets available in the LOD cloud, users need to be

familiar with formal languages used to query such da-

tasets (typically SPARQL), and should be aware of

the semantic model undelying the dataset, in order to

be able to formulate the proper query and to ”under-

stand” the resources and properties in the query re-

sults.

In order to face such difﬁculties, a number of

tools have been proposed, aimed at providing sim-

ple query interfaces to non-expert users – see, for ex-

ample, (Heim et al., 2010) (Russell and Smart, 2010)

(Haag et al., 2015) among many others.

In this perspective, a user-friendly approach to

query RDF datasets is keyword search; for example,

(Ouksili et al., 2016) presents an enhanced appro-

ach to search through RDF datasets using keywords:

Ouksili and colleagues’ approach is based on patterns

used to include external knowledge in the search pro-

cess in order to improve the result quality. Howe-

ver, the most promising approach to support users in

querying RDF repositories seems to be faceted search

(Tunkelang, 2009) (Tzitzikas et al., 2017). Facets are

predicate-value pairs (e.g., gender= female) and fa-

ceted search can be seen as an interactive model for

query formulation in which users can progressively

apply ﬁlters in the form of facets to narrow down the

results. There are several tools based on this appro-

ach, such as (Berners-Lee et al., 2008) (Hahn et al.,

2010) (Heim et al., 2008) (Hildebrand et al., 2006)

(Huynh and Karger, 2009), and more recently (Fafa-

lios and Tzitzikas, 2013) (Navarro et al., 2015) and

(Graub et al., 2016), where the authors present Sem-

Facet, a faceted search tool enhanced by the use of

ontological axioms for enriching results with impli-

cit information. Moreover, in this track, it is worth

mentioning the ontology-aware system by Hyv

onen

and colleagues (Kurki and Hyv

onen, 2010), and the

application of their approach in the cultural heritage

domain (Hyv

onen et al., 2005).

Another tool aimed at supporting users in LOD

exploration is PepeSearch (Vega-Gorgojo et al.,

2016), that provides a form-oriented user interface

based on the semantic model underlying the data-

set to be explored. Other similar tools are Optique-

VQS (Soylu et al., 2016) and Surveyor (Vega-Gorgojo

et al., 2017), a CORS-enabled SPARQL 1.1 endpoint

plugin available without installation, providing users

with a view of the content of the selected dataset and

enabling navigation through classes and instances.

One of the most challenging issue when exploring

LOD sets is the mapping between semantic models as

well as between entities across different datasets. As

far as semantic models alignment is concerned, the

relevant research area is ontology matching – see, for

instance, (Potoniec et al., 2017) – but it falls outside

the scope of the present paper. As far as the map-

ping between entities described in different datasets is

concerned, it is worth mentioning some well-known

entity linking (or instance matching) tools: LIMES,

a time-optimized tool for link discovery (Ngonga-

Ngomo and Auer, 2011) (Ngonga-Ngomo, 2012) and

SILK (Volz et al., 2009). We will see how such tools

can be relevant in our approach in Section 3.4.

3 LDscout

LDscout executes an Exploration Task on a given da-

taset, retrieving all data about a given set of instan-

ces, described using a given set of terms (vocabulary).

We will start by deﬁning the notions of Exploration

Task and Solution to an Exploration Task in general

(Section 3.1), then we will instantiate these notions

on the speciﬁc case of RDF datasets and OWL ontolo-

gies (Section 3.2). Since computing the solution to an

exploration task may be often unfeasible in practice,

we will then characterize speciﬁc subsets of the so-

lutions to RDF-OWL Exploration Tasks, which can

be easily computed and that still provide useful insig-

hts on data (Section 3.2); we will then show how this

sub-solutions can be actually computed by means of

SPARQL queries (Section 3.3).

Exploring RDF Datasets with LDscout

3.1 Formal Deﬁnition of an Exploration

Task

Deﬁnition 3.1. An Exploration Task is a triple ET =

hV, I, dsi, where V is a non-empty and possibly inﬁ-

nite set of terms, I is a non-empty and possibly inﬁnite

set of instances (where V ∩ I =

0) and ds is a dataset

(i.e. a set of data elements).

Intuitively, an exploration task represents the task

of retrieving from the dataset ds all and only those

data elements that describe the instances in I and that

are expressed by means of the terms in V . The con-

straint V ∩ I =

0 states that in any exploration task

there is a rigid distinction between the elements that

play the role of instances about which data are sear-

ched and those that play the role of terms used to des-

cribe those instances.

In a typical setting, I is the set of entities an agent

(being either a human user or a software application)

is focusing on and that (s)he/it is searching data about,

V is a vocabulary that the agent can understand (and

that denotes the set of notions the agent is currently

interested in) and ds is the dataset that is being explo-

red.

A solution to an exploration task is deﬁned as fol-

lows:

Deﬁnition 3.2. The solution to an Exploration Task

ET = hV, I, dsi is the largest set SOL(ET ) ⊆ ds, such

that each data element in SOL(ET ) is about a non-

empty set of instances INST ⊆ I and it describes the

instances in INST using only terms contained in V .

Intuitively, a solution to an exploration task is the

largest subset of the explored dataset ds that describes

entities speciﬁed in I, by means of terms in V .

From the deﬁnitions above, it immediately follows

that:

Proposition 3.1. Any exploration task always has one

and only one (possibly empty) solution.

For example, let us consider the following sets:

• V

= {Woman, Man, Factory, age, works at,

collaborates with}

• V

= {Person, knows}

• I

= {Mary, Alice, Bob}

• I

= {Betty, Bob}

• ds = {Woman(Mary), Woman(Alice),

Man(John), Man(Steve),

Consultant(John), Consultant(Alice),

age(Mary, 45), Company(ACME),

works at(Mary, ACME),

collaborates with(John,Mary),

collaborates with(Alice, Mary),

collaborates with(Steve, John)}

Given the sets speciﬁed above, the solution to the

exploration task ET

= hV

, I

, dsi is

SOL(ET

) = {Woman(Mary), Woman(Alice),

age(Mary, 45), works at(Mary, ACME),

collaborates with(John, Mary),

collaborates with(Alice, Mary)};

As regards the problem formulation, it is worth

noting that: Not all the terms employed in ds are men-

tioned in V

(e.g. Consultant 6∈ V

, but it is used in

ds); the same holds for instances (Steve 6∈ I

, but it is

described in ds); conversely, Factory belongs to V

but it does not belong to the language of ds and Bob

is included in I

, but it is not described in ds.

Moreover, for what it concerns the solution, it is

worth noting that: Man(John), Man(Steve),

collaborates with(Steve, John) 6∈ SOL(ET

even though Man, collaborates with ∈ V

and

this is due to the fact that John, Steve 6∈ I

;

Consultant(Alice) 6∈ SOL(ET

), even though

Alice ∈ I

and this is due to the fact that

Consultant 6∈ V

; works at(Mary, ACME),

collaborates with(John, Mary) ∈ SOL(ET

), even

though ACME, John 6∈ I

, and this is due to the fact

that works at, collaborates with ∈ V

and Mary ∈ I

As stated in Proposition 3.1, an exploration task

may have an empty solution, for instance: If ET

, I

, dsi, then SOL(ET

) =

0 (since no term in V

belongs to the language of ds); if ET

= hV

, I

, dsi,

then SOL(ET

) =

0 (since no instance in I

is descri-

bed in ds).

Among the exploration tasks, it is worth mentio-

ning those ones in which the vocabulary is the set V

of all possible terms and/or the set of instances is the

set I of all possible instances. Even though such tasks

do not present any formal peculiarity, they are con-

ceptually relevant, thus we spend a few words about

them.

Any exploration task of the form ET

hV , I, dsi (with I 6= I) speciﬁes the task of retrie-

ving from the dataset ds all the data about the instan-

ces in I, no matter the language in which such data

are expressed. For example, if ET

= hV , I

, dsi

and ds speciﬁed above), we have SOL(ET

) =

{Woman(Mary), Woman(Alice), Consultant(Alice),

works at(Mary, ACME),

collaborates with(Mary,John),

collaborates with(Alice, Mary)}

Any exploration task of the form ET

= hV, I, dsi

(with V 6= V ) represents the task of retrieving from

the dataset ds all the data expressed in terms of

the vocabulary V , no matter the instances they are

KMIS 2018 - 10th International Conference on Knowledge Management and Information Sharing

about. For example, if ET

= hV

, I, dsi (V

and ds speciﬁed above), we have SOL(ET

) =

{Woman(Mary), Woman(Alice), Man(John),

Man(Steve), works at(Mary, ACME),

collaborates with(Mary,John),

collaborates with(Alice, Mary),

collaborates with(Steve, John)}.

Any exploration task of the form ET

V ,I

hV , I, dsi represents the task of retrieving from the

dataset ds all the data about any instance, no matter

the language in which such data are expressed. For

example, if ET

V ,I

= hV , I, dsi (ds speciﬁed above),

we have SOL(ET

V ,I

) = ds.

3.2 RDF-OWL Exploration Tasks:

Exploration Tasks for RDF Datasets

that Use OWL Ontologies

The formal deﬁnitions stated in Section 3.1 can be

instantiated to a speciﬁc kind of exploration tasks,

which is relevant for both Semantic Web and Linked

Data: the case in which the dataset is an RDF data-

set, possibly using terms from OWL ontologies for

expressing (part of) its data.

RDF – Resource Description Framework – (W3C,

2014) is a model for data representation based on tri-

ples. A triple hx p yi represents a data element where

x is an RDF resource (i.e., anything that can be tal-

ked about), p is a property and y can be either a re-

source or a value. The meaning of a triple is that the

resource x is associated with y through the property

p (e.g., hMary works at ACMEi means that the re-

source Mary is associated with the resource ACME

through the property works at; hMary age 45i means

that the resource Mary is associated with the value 45

through the property age).

OWL – Web Ontology Language – (W3C, 2012)

is a formal language for representing ontologies in

a standard and machine-readable way. OWL allows

one to describe instances (e.g., Mary, ACME, etc.)

by grouping them in classes (e.g., Woman, Company,

etc.), by linking them through binary object proper-

ties (e.g., works at, collaborates with, etc.), by as-

sociating them with values through binary data pro-

perties (e.g., age, etc.). An OWL ontology (basically,

a set of axioms) can be seen as a formal description

of the semantics of a vocabulary whose terms refer to

It is worth noting that, in general, for an exploration

task ET

V ,I

= hV , I,dsi it can be SOL(ET

V ,I

) 6= ds, since all

the data not describing instances possibly present in ds (e.g.

the metadata about the terms of the language employed in

ds) should be ﬁltered out in the solution.

instances, classes, object properties, and data proper-

ties.

In the RDF-OWL setting, the deﬁnition of an ex-

ploration task is instantiated as follows:

Deﬁnition 3.3. An RDF-OWL Exploration Task is an

exploration task (Deﬁnition 3.1) RO-ET = hV = C ∪

OP ∪ DP ∪ {rdf: type, owl:sameAs}, I, dsi, where:

1. C is a possibly empty and possibly inﬁnite set of

OWL class names;

2. OP is a possibly empty and possibly inﬁnite set of

OWL object properties names;

3. DP is a possibly empty and possibly inﬁnite set of

OWL data properties names;

4. I is a non-empty and possibly inﬁnite set of RDF

resources, representing instances;

5. ds is an RDF dataset.

The deﬁnition above explicitly distinguishes the

different categories of (logical) terms by means of

which instances can be described in OWL (i.e., class,

object property and data property names). Moreo-

ver, the vocabulary always includes the pre-deﬁned

properties rdf: type and owl:sameAs, expressing the

membership of an instance to a class and the equiva-

lence between two instances, respectively.

The deﬁnition of the solution to an exploration

task (Deﬁnition 3.2) still holds for RDF-OWL Ex-

ploration Tasks. However, in many cases, solving an

RDF-OWL Exploration Task may be difﬁcult or even

unfeasible, in practice. The main source of difﬁculty

being that, in general, OWL data may be expressed by

means of complex class expressions, negative proper-

ties, and other OWL constructs, which make it difﬁ-

cult (if not impossible) to compute solutions to RDF-

OWL Exploration Tasks, especially if the dataset ds is

huge and only remotely accessible through SPARQL

queries.

For practical purposes, we thus deﬁne the notion

of Restricted Solution to an RDF-OWL Exploration

Task, as follows:

Deﬁnition 3.4. A Restricted Solution to an RDF-

OWL Exploration Task

RO-ET = hV = C ∪ OP ∪ DP ∪

{rdf:type, owl:sameAs}, I, dsi is the

set R-SOL(RO-ET) = R-SOL

(RO-ET) ∪

R-SOL

(RO-ET) ∪ R-SOL

(RO-ET) ∪

R-SOL

sameAs

(RO-ET), where:

1. R-SOL

(RO-ET) = {hi rdf:type ci ∈ ds/i ∈ I ∧

c ∈ C};

2. R-SOL

(RO-ET) = {hi op ji ∈ ds/op ∈ OP ∧

(i ∈ I ∨ j ∈ I)};

3. R-SOL

(RO-ET) = {hi d p vi ∈ ds/i ∈ I ∧ d p ∈

DP};

Exploring RDF Datasets with LDscout

4. R-SOL

sameAs

(RO-ET) = {hi owl:sameAs ji ∈

ds/i ∈ I ∨ j ∈ I};

The deﬁnition above takes into considera-

tion only those RDF triples that express clas-

ses (the R-SOL

(RO-ET) set), positive object

properties (R-SOL

(RO-ET)), positive data

properties (R-SOL

(RO-ET)) and identity

(R-SOL

sameAs

(RO-ET)) assertions on instances.

Moreover, the considered class assertions have the

form hi rdf: type ci, where c is a class name. There-

fore, the other kinds of OWL assertions on instances

that may be present in an RDF dataset (i.e., negative

properties, difference among individuals, and com-

plex class expression assertions on instances), are

ruled out.

The above-stated deﬁnitions immediately entail

the two following properties:

Proposition 3.2. Any RDF-OWL Exploration Task

always has one and only one (possibly empty) Re-

stricted Solution.

Proposition 3.3. If RO-ET = hV = C ∪ OP ∪

DP, I, dsi is an RDF-OWL Exploration Task, it holds

that R-SOL(RO-ET) ⊆ SOL(RO-ET).

Intuitively, Proposition 3.3 states that every re-

stricted solution to an RDF-OWL exploration task is

also a solution to it.

As an example, let us consider the dataset ds

in Section 3.1. If we interpret the unary pre-

dicates as class names, the predicate age as a

data property name, and the other predicates as

object property names, we can easily rewrite ds

into an RDF dataset ds

RDF

, where, for exam-

ple, “Woman(Mary)” and ”age(Mary,45)” become

“hMary rdf:type Womani” and “hMary age 45i”

respectively). If C

= {Woman, Man, Factory},

= {works at, collaborates with}, DP

= {age},

= C

∪ OP

∪ DP

∪ {rdf: type, owl:sameAs} and

= {Mary, Alice, Bob}, we can specify the RDF-

OWL Exploration Task RO-ET

= hV

, I

, ds

RDF

The restricted solution R-SOL(RO-ET

) to this ex-

ploration task is the union of the following sets:

1. R-SOL

(RO-ET

) = {hMary rdf: type Womani,

hAlice rdf:type Womani};

2. R-SOL

(RO-ET

) = {hMary works at ACMEi,

hAlice collaborates with Maryi,

hJohn collaborates with Maryi};

3. R-SOL

(RO-ET

) = {hMary age 45i};

4. R-SOL

sameAs

(RO-ET

) =

In this speciﬁc case, we have R-SOL(RO-ET

) =

SOL(RO-ET

) (i.e. the restricted solution is also the

solution).

3.3 Computing Restricted Solutions to

RDF-OWL Exploration Tasks

For practical purposes, in the following, we consider

only RDF-OWL Exploration Tasks RO-ET = hV =

C ∪ OP ∪ DP ∪ {rdf: type, owl:sameAs}, I, dsi in

which the sets C, OP, DP and I are either ﬁnite or

they are, respectively, the set C of all possible class

names, OP of all possible object property names, DP

of all possible data property names and I of all possi-

ble instances.

The Restricted Solution to RO-ET can be compu-

ted by means of SPARQL queries.

SPARQL – SPARQL Protocol and RDF Query

Language – (W3C, 2013) offers a language to

query RDF datasets. The core part of a SPARQL

query is represented by a set of triple patterns

that express the main conditions that the pieces

of data must meet. Each query execution matches

these patterns with the data in order to single

out all the possible pattern instantiations. For in-

stance, if we consider the above-speciﬁed dataset

ds, the triple patterns [h?x collaborates with ?yi,

h?y works at ACMEi]

can be instantiated in the two

following ways: [hJohn collaborates with Maryi,

hMary works at ACME] and

[hAlice collaborates with Maryi,

hMary works at ACMEi]. Pattern instanti-

ations can be restricted by means of suit-

able ﬁlters. For instance, the SPARQL ﬁlter

“FILT ER EX IST S{?x rdf: type Woman}” would

retain only the latter instantiation. SPARQL offers

four query forms, including ASK and SELECT. The

former simply tests whether the queried dataset

provides at least one instantiation for the speciﬁed

set of triple patterns (satisfying the optional ﬁlters);

the latter returns as input a (combination of a) subset

of the variable bindings in the pattern instantiations

(that satisfy the optional ﬁlters).

In order to specify how this computation can be

done, we consider the restricted solutions’ subsets lis-

ted in Deﬁnition 3.4 and show how to calculate them;

with RES(Q) we indicate the result of a SPARQL

query Q.

The set R-SOL

(RO-ET) of class assertions is

computed as follows:

R-SOL

(RO-ET) =

•

0, if C =

• {hi rdf:type ci/i ∈ I ∧ c ∈ C ∧

RES(ASK W HERE{i a c.}) == true},

Elements preceded by a question mark “?” denote va-

riables.

KMIS 2018 - 10th International Conference on Knowledge Management and Information Sharing

if C 6=

0 ∧C 6= C ∧ I 6= I;

• {h?x rdf:type ci/c ∈ C∧?x ∈

RES(SELECT ?x W HERE {?x a c.}),

if C 6=

0 ∧C 6= C ∧ I = I;

• {hi rdf:type ?yi/i ∈ I∧?y ∈

RES(SELECT ?y W HERE {i a ?y.}),

if C = C ∧ I 6= I;

• {h?x rdf: type ?yi/h?x ?yi ∈ RES(SELECT ?x ?y

W HERE {?x a ?y.

FILT ER (EXIST S {?x a owl: NamedIndividual} ||

EXIST S {?y a owl:Class})},

if C = C ∧ I = I.

Starting from the ﬁrst case above, if no class name

is speciﬁed (i.e., C =

0), no class assertion is retrie-

ved from the dataset. If non-empty and ﬁnite sets of

class names and instances are speciﬁed (C 6=

0 ∧C 6=

C ∧ I 6= I) an ASK SPARQL query for each pair in

I ×C is performed to check whether the correspon-

ding class assertion is present in the dataset. If a non-

empty and ﬁnite set of class names is speciﬁed for

any possible instance (C 6=

0 ∧C 6= C ∧ I = I), all the

class assertions involving a speciﬁed class name is re-

trieved, no matter the instance it refers to. If the set

of all possible class names is speciﬁed for a ﬁnite set

of instances (C = C ∧ I 6= I), all the class assertions

on the speciﬁed instances are retrieved, no matter the

employed class name. If the sets of all possible class

names and instances are speciﬁed (C = C ∧ I = I), all

the class assertions on instances present in the data-

set are retrieved. In this last case, it should be noted

that all the class assertions possibly present in a RDF-

OWL dataset that do not refer to instances (i.e. to

OWL named individuals) should be ruled out. This

is the case, for example, of those assertions that spe-

cify that an ontology element is an owl : Class, an

owl : DatatypeProperty, etc. Here we assume that

the dataset explicitly speciﬁes the OWL type at least

for the individuals it describes or for the classes it em-

ploys. If this is the case, we can specify a simple ﬁlter

that rules out unwanted triples (otherwise, we would

need more elaborated ﬁlters).

The set R-SOL

(RO-ET) of object property as-

sertions is computed as follows:

R-SOL

(RO-ET) =

•

0, if OP =

• {hi op ?yi/i ∈ I ∧ op ∈ OP∧ ?y ∈

RES(SELECT ?y W HERE {i op ?y.})}

∪

{h?x op ii/i ∈ I ∧ op ∈ OP∧ ?x ∈

RES(SELECT ?x W HERE {?x op i.})},

if OP 6=

0 ∧ OP 6= OP ∧ I 6= I;

• {h?x op ?yi/op ∈ OP ∧ h?x ?yi ∈

RES(SELECT ?x ?y W HERE {?x op ?y.})},

if OP 6=

0 ∧ OP 6= OP ∧ I = I;

• {hi ?p ?yi/i ∈ I ∧ h?p ?yi ∈

RES(SELECT ?p ?y W HERE {i ?p ?y.})}

∪

{h?x ?p ii/i ∈ I ∧ h?x ?pi ∈

RES(SELECT ?x ?p W HERE {?x ?p i.})},

if OP = OP ∧ I 6= I;

• {h?x ?p ?yi/h?x ?p ?yi ∈

RES(SELECT ?x ?p ?y W HERE {?x ?p ?y.

FILT ER (EXIST S {?x a owl: NamedIndividual}||

EXIST S {?p a owl:ObjectProperty}||

EXIST S {?y a owl:NamedIndividual})},

if OP = OP ∧ I = I;

If no object property name is speciﬁed (i.e. OP =

0), no object property assertion is retrieved from the

dataset. If non-empty and ﬁnite sets of object property

names and instances are speciﬁed (OP 6=

0 ∧ OP 6=

OP ∧ I 6= I), then each triple that represents an object

property assertion through a speciﬁed object property

and that is about a speciﬁed instance (which can be

either subject or object of the triple) is retrieved. If a

non-empty and ﬁnite set of object property names is

speciﬁed for any possible instance (OP 6=

0 ∧ OP 6=

OP ∧ I = I), then all the object property assertions in-

volving a speciﬁed object property name is retrieved,

no matter the instances it refers to. If the set of al

possible object property names is speciﬁed for a ﬁ-

nite set of instances (OP = OP ∧ I 6= I), all the object

property assertions on the speciﬁed instances are re-

trieved, no matter the employed object property name

and the subject or object role of the speciﬁed instances

in the triples. If the sets of all possible object property

names and instances are speciﬁed (OP = OP ∧ I = I),

all the object property assertions on instances present

in the dataset are retrieved. As in the case of class as-

sertions, also in this case suitable ﬁlters are speciﬁed

to rule out unwanted triples.

Sets R-SOL

(RO-ET) – data property assertions

– and R-SOL

sameAs

(RO-ET) – identity assertions on

instances – mentioned in Deﬁnition 3.4, are computed

in a similar way.

3.4 The Role of LDscout in PRiSMHA

Figure 1 shows an overview of PRiSMHA architec-

ture, focusing on the role that LDscout plays in it.

We only provide here a very quick sketch of this ar-

chitecture, in order to describe the way we are using

LDscout in the project: a detailed description of the

other mentioned modules is out of the scope of this

paper and can be found in (Caserio et al., 2017).

Exploring RDF Datasets with LDscout

The system offers a crowd-sourcing platform ena-

bling users to specify ontology-based formal seman-

tic representations of the content of historical docu-

ments. Following a quite standard approach – e.g.,

(Cybulska and Vossen, 2011), (Sprugnoli and Tonelli,

2016) – such semantic representations are centered on

the notion of event. Basically, events are described by

means of their types, the places where they occur, the

time when they happen and the entities that participate

in them. These representations are based on the His-

torical Event Representation Ontology (HERO) (Ca-

serio et al., 2017) and are stored in an RDF triplestore.

The crowd-sourcing platform relies on the support

provided by the Information Extraction module and

the LOD mining module. The Information Extraction

module analyzes texts (when available) extracting in-

formation about entities (people, organizations, pla-

ces), which are used to support users in the identiﬁ-

cation of relevant events and participants within the

documents (Rovera et al., 2017).

Figure 1: The role of LDscout in PRiSMHA architecture.

The LOD mining module queries LOD datasets

in order to get more information about the entities

identiﬁed by users in the documents. For example,

if the user is reading a text about CGIL (the Italian

General Confederation of Labour), and needs some

more information about it, the LOD mining module

can get such information from DBpedia (dbpedia.org)

and provide it to the user. The retrieved information

(e.g., that the acronym means ”Confederazione Ge-

nerale Italiana del Lavoro”, that it is an Italian Trade

Union, and it was founded in June 1944, etc.) is re-

presented as a set of RDF triples, expressed in terms

of the HERO vocabulary, so that it can be immedia-

tely used to enrich the semantic representation under

construction on the crowd-sourcing platform.

In order to query LOD datasets and get results

immediately usable on the PRiSMHA platform, the

LOD mining module exploits LDscout APIs. For

conﬁguring the suitable Exploration Task in such a

way that ensures results (Exploration Task Soluti-

ons) in terms of HERO, the LOD mining module re-

lies on a set of mappings between: (a) HERO clas-

ses/properties and the vocabulary used in the selected

dataset; (b) instances in the PRiSMHA triplestore and

instances in the selected dataset. In the current proof-

of-concept prototype, such mappings ensure intero-

perability with DBpedia and they have been manu-

ally encoded. However, if ontological mappings (i.e.,

those concerning classes and properties) can be seen

as part of the design and implementation of the onto-

logy interoperability, for instance mappings automa-

tic tools such as LIMES (see Section 2) should be ta-

ken into account, in order to guarantee the scalability

of the approach.

The RDF dataset produced by the crowd-sourcing

platform is accessible through a standard SPARQL

endpoint. However, in order to hide SPARQL com-

plexity, providing a much easier to use interface, the

dataset can also be accessed through a RESTful ser-

vice, implemented thanks to LDscout APIs. In this

case, the Exploration Task is deﬁned on the dataset

containing the semantic representation produced by

the platform, enabling software clients to easily conﬁ-

gure their queries with different sets of HERO classes

and properties, and different sets of instances.

The availability of a RESTful service to access the

semantic representations greatly simpliﬁes the exploi-

tation of the PRiSMHA dataset by third-party appli-

cations that aims at exploiting our data for different

purposes, ranging from history-aware tourist guides

to education-oriented tools.

4 CONCLUSIONS

In this paper, we presented a general model handling

exploration tasks over datasets and we described its

instantiation in LDscout, a Java library for querying

RDF triplestores by specifying the vocabulary (onto-

logy) and the set of instances the client is interested

KMIS 2018 - 10th International Conference on Knowledge Management and Information Sharing

in.

To assess the usefulness of LDscout, we also sho-

wed how we used it within the PRiSMHA platform.

With respect to this usage, the ﬁrst improvement we

are working on concerns the mappings between in-

stances in our triplestore and instances in LOD sets:

we are investigating the exploitation of automatic

instance mapping tools, such as LIMES (Ngonga-

Ngomo and Auer, 2011) in order to overcome the

work overload of manually encoding such mappings.

Moreover, an interesting enhancement is represen-

ted by endowing LDscout with a friendly web-based

user interface, available for human users, and not only

for software clients.

ACKNOWLEDGEMENTS

This work has been partially supported by Compag-

nia di San Paolo and Universit

a di Torino within the

PRiSMHA project.

REFERENCES

Alexiev, V. (2016). Linked Open Data for cultural heritage

and digital humanities. Ontotext Blog, September 28.

Berners-Lee, T., Hollenbach, J., Lu, K., Presbrey, J., Prud-

hommeaux, E., and Schraefel, M. M. C. (2008). Ta-

bulator redux: Browsing and writing linked data. In

Proc. LDOW 2008.

Caserio, M., Goy, A., and Magro, D. (2017). Smart access

to historical archives based on rich semantic metadata.

In Proc. KMIS 2017, pages 93–100. SciTePress.

Cybulska, A. and Vossen, P. (2011). Historical event ex-

traction from text. In Proc. LaTeCH 2011, pages 39–

43.

Edelstein, J., Galla, L., Li-Madeo, C., Marden, J., Rho-

nemus, A., and Whysel, N. (2013). Linked Open

Data for cultural heritage: Evolution of an information

technology. In Proc. 31st ACM International Confe-

rence on Design of Communication. Association for

Machine Computing.

Fafalios, P. and Tzitzikas, Y. (2013). X-ENS: Semantic en-

richment of web search results at real-time. In Proc.

SIGIR 2013, pages 1089–1090.

Goy, A., Damiano, R., Loreto, F., Magro, D., Musso, S.,

Radicioni, D., Accornero, C., Colla, D., Lieto, A.,

Mensa, E., Rovera, M., Astrologo, D., Boniolo, B.,

and D’ambrosio, M. (2017). Prismha (providing rich

semantic metadata for historical archives). In Proc.

Contextual Representation of Objects and Events in

Language (CREOL 2017).

Goy, A., Magro, D., and Rovera, M. (2015). Ontologies and

historical archives: A way to tell new stories. Applied

Ontology, 10(3-4):331–338.

Graub, M. A. N. B. C., Kharlamovb, E., Marciu

ska, S., and

Zheleznyakov, D. (2016). Faceted search over RDF-

based knowledge graphs. Web Semantics: Science,

Services and Agents on the World Wide Web, 37-

38:55–74.

Haag, F., Lohmann, S., Siek, S., and Ertl, T. (2015). Visual

querying of linked data with QueryVOWL. In Joint

Proc. SumPre 2015 and HSWI 2014-2015. CEUR.

Hahn, R., Bizer, C., Sahnwaldt, C., Herta, C., Robinson,

S., B

urgle, M., D

uwiger, H., and Scheel, U. (2010).

Faceted Wikipedia search. In Proc. BIS 2010, pages

1–11.

Heath, T. and Bizer, C. (2011). Linked Data: Evolving the

Web into a Global Data Space. Morgan & Claypool.

Heim, P., Ertl, T., and Ziegler, J. (2010). Facet graphs:

Complex semantic querying made easy. In Proc.

ESWC 2010, pages 288–302.

Heim, P., Ziegler, J., and Lohmann, S. (2008). gfacet: A

browser for the web of data. In IMC-SSW 2008, pages

49–58.

Hildebrand, M., van Ossenbruggen, J., and Hardman, L.

(2006). /facet: A browser for heterogeneous semantic

web repositories. In Proc. ISWC 2006, pages 272–

285.

Huynh, D. F. and Karger, D. R. (2009). Parallax and com-

panion: Set-based browsing for the data web. In Proc.

WWW 2009. ACM.

Hyv

onen, E., M

akel

a, E., Salminen, M., Valo, A., Viljanen,

K., Saarela, S., Junnila, M., and Kettula, S. (2005).

Museumﬁnland - ﬁnnish museums on the semantic

web. Journal of Web Semantics, 3(2-3):224–241.

Kurki, J. and Hyv

onen, E. (2010). Collaborative metadata

editor integrated with ontology services and faceted

portals. In Proc. ORES 2010.

Navarro, J. F. G., Villamar, V. A. L., Srinivasan, J., Perry,

M., Das, S., and Wu, Z. (2015). Exploring large RDF

datasets using a faceted search. In Proc. ISWC 2015 -

Posters and Demos. CEUR.

Ngonga-Ngomo, A. C. (2012). On link discovery using a

hybrid approach. Journal on Data Semantics, 1:203–

217.

Ngonga-Ngomo, A. C. and Auer, S. (2011). LIMES - a

time-efﬁcient approach for large-scale link discovery

on theweb of data. In Proc. IJCAI 2011, pages 2312–

2317.

Ngonga-Ngomo, A. C., Auer, S., Lehmann, J., and Zaveri,

A. (2014). Introduction to linked data and its lifecycle

on the web. In Koubarakis, M., Stamou, G., Stoilos,

G., Horrocks, I., Kolaitis, P., Lausen, G., and Weikum,

G., editors, Proc. Reasoning Web 2014, LNCS 8714,

pages 1–99. Springer.

Ouksili, H., Kedad, Z., Lopes, S., and Nugier, S. (2016).

Pattern-based keyword search on RDF data. In Proc.

ESWC Satellite Events 2016, pages 30–34.

Potoniec, J., Jakubowski, P., and Ławrynowicz, A. (2017).

Swift linked data miner: Mining OWL 2 EL class ex-

pressions directly from online RDF datasets. Web Se-

mantics: Science, Services and Agents on the World

Wide Web, 46-47:31–50.

Exploring RDF Datasets with LDscout

Rovera, M., Nanni, Ponzetto, S. P., and Goy, A. (2017).

Domain-speciﬁc named entity disambiguation in his-

torical memoirs. In Proc. CLiC-it 2017. CEUR.

Russell, A. and Smart, P. R. (2010). NITELIGHT: A graphi-

cal editor for SPARQL queries. In Proc. ISWC 2008 -

Posters and Demos.

Soylu, A., Giese, M., Jimenez-Ruiz, E., Vega-Gorgojo, G.,

and Horrocks, I. (2016). Experiencing optiquevqs: a

multi-paradigm and ontology-based visual query sy-

stem for end users. Universal Access in the Informa-

tion Society, 15(1):129–152.

Sprugnoli, R. and Tonelli, S. (2016). One, no one and

one hundred thusand events: Deﬁning and processing

events in an inter-disciplinary perspective. Natural

Language Engineering, 23(4):485–506.

Tunkelang, D. (2009). Faceted Search, Synthesis Lectu-

res on Information Concepts, Retrieval, and Services.

Morgan & Claypool.

Tzitzikas, Y., Manolis, N., and Papadakos, P. (2017). Face-

ted exploration of RDF/S datasets: A survey. Journal

of Intelligent Information Systems, 48(2):329–364.

Vega-Gorgojo, G., Giese, M., and Slaughter, L. (2017). Ex-

ploring semantic datasets with RDF surveyor. In Proc.

ISWC 2017 - Posters and Demos and Industry Tracks.

Vega-Gorgojo, G., Slaughter, L., Giese, M., Heggestøyl, S.,

uwer, J. W., and Waaler, A. (2016). Pepesearch:

Easy to use and easy to install semantic data search.

In Proc. ESWC 2016 Satellite Events, pages 146–150.

Volz, J., Bizer, C., Gaedke, M., and Kobilarov, G. (2009).

Discovering and maintaining links on the web of data.

In Proc. ISWC 2009, pages 650–665.

W3C (2012). OWL 2 Web Ontology Language primer (se-

cond edition).

W3C (2013). SPARQL 1.1 overview.

W3C (2014). RDF 1.1 primer.

KMIS 2018 - 10th International Conference on Knowledge Management and Information Sharing

100