RecLAK: Analysis and Recommendation of
Interlinking Datasets
Giseli Rabello Lopes Luiz André P. Paes Leme
Departament of Informatics, PUC-Rio Computer Science Institute, UFF
Rio de Janeiro/RJ, Brazil Niterói/RJ, Brazil
grlopes@inf.puc-rio.br lapaesleme@ic.uff.br
Bernardo Pereira Nunes Marco A. Casanova
Departament of Informatics, PUC-Rio Departament of Informatics, PUC-Rio
Rio de Janeiro/RJ, Brazil Rio de Janeiro/RJ, Brazil
bnunes@inf.puc-rio.br casanova@inf.puc-rio.br
ABSTRACT collection of open data available on the Web. One of the
This paper presents the RecLAK , a Web application devel- datasets covered by the LinkedUp project is the Learning
oped for the LAK Challenge 2014. RecLAK focuses on the Analytics and Knowledge (LAK) dataset. The LAK dataset,
analysis of the LAK dataset metadata and provides recom- referred to as lak, provides access to structured fulltext and
mendations of potential candidate datasets to be interlinked metadata from key research publications in the field of learn-
with the LAK dataset. RecLAK follows an approach to gen- ing analytics and educational data mining6 . lak is regu-
erate recommendations based on Bayesian classifiers and on larly updated with data, for instance, from the LAK (Learn-
Social Networks Analysis measures. Furthermore, RecLAK ing Analytics and Knowledge) and EDM (Educational Data
generates graph visualizations that explore the LAK dataset Mining) conference series. According to the DataHub meta-
over other datasets in the Linked Open Data cloud. The re- data, lak was not linked to other datasets, except DBpedia.
sults of the experiments contribute to the understanding and However, an exploratory search in the DataHub in fact re-
improvement of the LAK dataset. Furthermore, it can also vealed related datasets that lak could be linked to, such as
help researchers of the fields covered by LAK dataset, such other bibliographic datasets.
as learning analytics and educational data mining.
This scenario is very common. Most of the published datasets
are still awaiting to be linked and, therefore, they do not ful-
1. INTRODUCTION fill the requirements to be considered 5-star [1] and fail to
The effort of publishing Linked Data has been accompanied take advantage of other data. Basically, as argued in [10],
by the creation of catalogs of Linked Data datasets, such as the linkage to popular datasets is favoured for two main
the DataHub 1 , to make data findable and reusable. How- reasons: the difficulty of finding related open datasets; and
ever, despite the fact that extensive lists of open datasets the strenuous task of discovering instance mappings between
are available in these catalogs, most of the data publishers different datasets.
typically link their datasets only to popular ones, such as
DBpedia2 , Freebase3 and Geonames4 . Although the link- In this sense, lak will be explored as a case study. The rec-
age to popular datasets allows the exploration of external ommendation challenge associated to the interlinking of lak
resources, it fails to cover more specialized data. in the LOD can be posed by considering two main questions:
As a practical example of this scenario, we may highlight
the LinkedUp project5 , which is an initiative that aims at Q1. For a dataset d, published in the LOD, is it interesting
providing educational organizations and institutions with a for the publisher of d to try to link it to lak ?
1
http://datahub.io/
2
http://dbpedia.org Q2. For a dataset d, published in the LOD, is it interesting
3
http://www.freebase.com for the lak administrator to try to link his dataset to
4
http://www.geonames.org/ d?
5
http://linkedup-project.eu
In more detail, let t and di be two datasets. A link from t
to di is a triple of the form (s, p, o) such that s is defined
in t and o is defined in di . We say that t is linked to di , or
that di is linked from t, iff there is at least a link from t to
di . We also say that di is relevant for t iff there is at least
a resource defined in di that can be linked from a resource
defined in t.
6
http://lak.linkededucation.org
Questions Q1 and Q2 are special cases of the dataset inter-
linking recommendation problem posed as follows: X
!
score(di , t) = log(P (Fj |Di )) + log(P (Di )) (1)
j=1..n
Given a finite set of datasets D and a dataset
t, compute a rank score for each dataset di ∈ D Based on the maximum likelihood estimate of the probabil-
such that the rank score of di increases with the ities [8] in a training set of datasets, the above probabilities
chances of di being relevant for t. can be estimated as follows:
In this paper, we first introduce two rank score functions to count(fj , di ) count(di )
P (Fj |Di ) = Pn ; P (Di ) = Pm
address the dataset interlinking recommendation problem. j=1 count(f j , d i ) i=1 count(di )
Then, we apply the functions to answer question Q2.
where count(fj , di ) is the number of datasets in the train-
The remainder of this paper is organized as follows. Sec- ing set that have feature fj and that are linked to di , and
tion 2 presents related work. Section 3 briefly describes count(di ) is the number of datasets in the training set that
the recommendation approaches. Section 4 shows the re- are linked to di , disregarding the feature set.
sult analysis of the metadata exploration and the generated
recommendations. Finally, Section 5 presents some final re- For the score function computation, some auxiliary functions
marks. help to avoid computing log(0) replacing this value by c,
which is a constant small enough to penalize the datasets di
2. RELATED WORK that do not have datasets with features Fj linked to them
In this paper, we use an extended version [5] of previous work or that do not have links from other datasets [5]. Thus, the
[3, 4], that introduced the rank score functions based on the idea is that, if the set of features of t is very often correlated
Bayesian and the Social Network approaches. The extended with datasets that are linked to di and t is not already linked
version also explores different sets of features related to the to di , then it is recommended to try to link t to di .
metadata of the datasets, such as properties, classes and
vocabularies, to compute the rank score functions. 3.2 Social Network-based ranking
We propose to analyze the dataset interlinking recommen-
Nikolov et al. [9, 10] propose an approach to identify rel- dation problem in much the same way as the link prediction
evant datasets for interlinking applying keywords searches problem in Social Networks [7]. Analogously, the Linked
and ontology matching techniques. Kuznetsov [2] describes Data network for D is a directed graph such that the nodes
a linking system, which is responsible for discovering rele- are the datasets in D and there is an edge between datasets
vant datasets for a given dataset and for creating instance u and v in D iff there is a link from u to v. To obtain more
level linkage. When compared with these approaches, the accurate results, we combine two measures, Preferential At-
rank score functions applied in this paper use only meta- tachment (pa) and Resource Allocation (ra), into a single
data and are, therefore, much simpler to compute and yet score [5], defined as follows:
achieve a good performance [5].
Lóscio et al. [6] and Wagner et al. [15] propose techniques to pa(t, di )
score(t, di ) = ra(t, di ) + (2)
find relevant datasets for user queries. The first approach is |D|
based on information quality criteria of correctness, schema
completeness and data completeness while the second one X 1
is based on the overlapping of sets of instances of datasets. pa(t, di ) = |Pdi | ; ra(t, di ) =
dj ∈St ∩Pd
Pdj
Oliveira et al. [13] use application queries and user feedback i
to discover relevant datasets. These papers aim at recom-
mending datasets with respect to user queries, which is a where Pdi is the popularity set of a dataset di ∈ D, that
problem close, but not identical to the problem discussed in is, the set of all datasets in D that have links to di , and
this paper. St is the similarity set of a dataset t, that is, the set of all
datasets in D that have features in common with t.
Nunes et al. [11, 12] performed several analysis on lak but
their focus was mainly in the dataset content. They also pro- The combined score induces the ranking of the datasets in D
posed other datasets to be interlinked with lak considering (from the largest to the smallest score) and gives priority to
their links with DBpedia. By contrast, this paper focuses on the ra score; the pa score, normalized by the total number
analyzing the metadata for creating rankings of candidate of datasets to be ranked (|D|), will play a role when there
datasets to be interlinked with lak using different recom- is a tie or when the ra value is zero.
mendation techniques.
4. RESULT ANALYSIS
3. RECOMMENDATION APPROACHES 4.1 Data used in the experiments
3.1 Bayesian ranking We selected a subset of the datasets indexed by the DataHub,
A rank score function, inspired on conditional probabilities, using the Learning Analytics and Knowledge dataset [14] as
that induces the ranking of the datasets in D (from the the target of the recommendation. From the DataHub cat-
largest to the smallest score), can be defined as follows: alog, we managed to obtain 295 datasets with at least one
21/02/14 12
transparency-linked-data
ecb-linked-data bfs-linked-data
eu-who-is-who
ontos-news-portal
eu-institutions grrp
oecd-linked-data world-bank-linked-data
eu-parliament-media the-eurostat-linked-data
red-uno-internacional-santillana
istat-immigration
global-hunger-index-2011
euskadi-farmacias
lak
b3kat eprtr norwegian-geo-divisions rechtspraak
educationalprograms_sisvu
lobid-resources dblp-deusto-gnoss
eea-rod open-data-risp
sandrart-net eurostat-rdf
interactivemaps-gnoss
rkb-explorer-darmstadt
national-diet-library-authorities enakting-energy
museosespania-gnoss
rkb-explorer-kaunas
proyectoapadrina
rkb-explorer-jisc lobid-organisations prospects-and-trends-gnoss
rkb-explorer-nsf farmers-markets-geographic-data-united-states nextweb-gnoss
rkb-explorer-ibm
rkb-explorer-risks
rkb-explorer-epsrc gnoss
knoesis-linked-sensor-data
nytimes-linked-open-data
rkb-explorer-ieee didactalia
geonames-semantic-web
rkb-explorer-pisa rkb-explorer-laas
rkb-explorer-southampton telegraphis uk-legislation-api
rkb-explorer-roma ineverycrea
rkb-explorer-wiki environment-agency-bathing-water-quality
miguiadeviajes-gnoss
garnicaplywood
rkb-explorer-newcastle enakting-population
event-media
rkb-explorer-deploy
rkb-explorer-eprints
rkb-explorer-kistirkb-explorer-irit rkb-explorer-ft
aegp-spanish-textile-and-clothing-association
artenuevosmedios-gnoss
deustoentrepreneurship
rkb-explorer-rae2001 fao-linked-data
rkb-explorer-resexrkb-explorer-citeseer
rkb-explorer-dblp green-competitiveness-gnoss ineverycrea-argentina
ordnance-survey-linked-data
rkb-explorer-acm rkb-explorer-eurecom
rkb-explorer-lisbon open-data-euskadi
chronicling-america
rkb-explorer-curriculum museums-in-italy
rkb-explorer-ulm rkb-explorer-budapest rkb-explorer-ecs
rkb-explorer-cordis biographical-directory-of-the-united-states-congress idreffr japan-radioactivity-stat
southampton-ecs-eprints
rkb-explorer-dotac ecs
my-experiment hellenic-police
geowordnet
rkb-explorer-courseware
rkb-explorer-italy jamendo-dbtune
psh-subject-headings hellenic-fire-brigade
dnb-gemeinsame-normdatei john-goodwins-family-tree sweto-dblp
sudocfr
lcsh rkb-explorer-deepblue rkb-explorer-unlocode
geospecies enakting-mortality
eunis semanticweb-org enakting-crime
diavgeia dcs-sheffield
national-diet-library-subject-headings
icane rkb-explorer-wordnet
hebis-bibliographic-resources enakting-nhs
rkb-explorer-webconf
w3c-wordnet agris
msc lexvo
rkb-explorer-os
stw-thesaurus-for-economics sztaki-lod fao-geopolitical-ontology
oclc-fast
glottolog-langdoc
linked-open-camera
deutsche-biographie
taxonconcept tags2con-delicious
gesis-thesoz asjp
Figure 1: The datasets and their links.
feature (class, property or vocabulary). Among the datasets and that are used in many datasets, such as owl:sameAs,
with links defined, there are 139 datasets with 697 known rdf:Property, rdfs:Resource, among others. The core of the
links. Figure 1 presents a graph representing the datasets selected set comes from the SWC ontology7 (Semantic Web
and their known links. In this graph, the size of a dataset Conference), which describes academic conferences and es-
node is proportional to the number of datasets linked to it tablishes a convention on how to use classes and properties
(in-degree). from other ontologies, mostly FOAF (Friend of a Friend ),
for people and organizations, and SWRC (Semantic Web for
The number of distinct features between classes and proper- Research Communities), for papers. It also includes meta-
ties was 11,868. The number of relations between datasets data from other ontologies, such as SIOC (Semantically-
and classes or properties was 16,750, where 6,447 were refer- Interlinked Online Communities) and DC (Dublin Core).
ences to classes and 10,303 were references to properties. For The selected lak features added to 37, where 31 of them
the details on how we extracted metadata from the DataHub are shared by other datasets in our set of data. A preview
catalog, see [5]. of the RecLAK interface showing the selected lak classes is
presented in Figure 2.
4.2 LAK features 4.3 Datasets with LAK features
As features of lak, we used a selected set of classes and
The set of datasets (represented by their id in DataHub)
properties obtained from the lak and from the LinkedUp
that have at least one feature in common with lak consists
project Web site. We filtered out, from 51 initial features,
7
those that were not related to the content of the dataset http://data.semanticweb.org/ns/swc/ontology
RecLAK interface presenting the recommendations for LAK
is presented in Figure 4.
The top 10 recommendations generated by each of the two
approaches (Bayesian and Social Network-based rankings)
and the respective score values estimated for each recom-
mended dataset are presented in Table 2. The top 10 ranked
datasets for each approach will be briefly described below.
Bayesian ranking. The topmost-ranked is a generic
dataset with concepts from the Semantic Web community.
Dataset #2 is a well-known lexical database of English.
Datasets from #3 to #6 positions of the Bayesian ranking
presented tied scores. Dataset #3 is a dataset with concepts
from tags generated by human annotators. Dataset #4 de-
scribes people, research groups and publications of the mem-
bers of the Computer Science Department at the University
Figure 2: Preview of the RecLAK interface showing of Sheffield. Dataset #5 is maintained by the chamber of
the selected lak classes. deputies in Italy, which is working to publish quality linked
data in several domains, including research. Dataset #6
describes the DBLP digital library, which provides biblio-
Table 1: Top 10 datasets sharing features with lak. graphic information on major computer science journals and
Dataset id # shared features proceedings. dblp also indexes the papers published in the
rkb-explorer-webconf 31 LAK and EDM conferences. Dataset #7 is the Geonames
linked-open-vocabularies-lov 8 dataset, which contains information about geographical lo-
krystian-pietruszka 7 cations. Dataset #8 contains information about languages,
aksworg 7 words, characters, and other human language-related enti-
dcs-sheffield 6 ties to the Linked Data Web and Semantic Web. lexvo has
southampton-ac-uk-profile 6 links to WordNet and thesauris. Dataset #9 is a Linked
jamendo-dbtune 6 Data version of the Association for Computing Machinery
sudocfr 6 (ACM) digital library. Finally, dataset #10 is a dataset of
rkb-explorer-webscience 6 the Library of Congress Subject Headings (LCSH), which
msc 6 catalogs materials stored by the Library of Congress and
other libraries around the United States.
of 132 datasets, with 376 associations between datasets and Social Network-based ranking. Since, there is some
lak features. Figure 3 presents a graph representing the overlap between the top 10 recommendations of Social
datasets and their associated lak features. In this graph, Network-based (SN-based) and Bayesian ranking, we will
the size of a feature node is proportional to the number of comment the top 10 datasets ranked only by the SN-based
datasets having it. approach. Dataset #2 publishes the news vocabularies used
by The New York Times as Linked Open Data. It cov-
Among the lak features, the most popular are from ers data and resources about people, locations and orga-
DC: dc:title, shared by 60 datasets, and dc:creator, with nizations. Dataset #3 covers topics related to innovation,
56 datasets references, and from FOAF: foaf:name and technology, business and education. Dataset #6 has links
foaf:homepage with, respectively, 41 and 36 other datasets catalogued in the DataHub for other bibliographic datasets
beyond lak referring to them. The least popular features are such as Citeseer, DBLP, ACM, IEEE and EPrints. Dataset
metadata directly from SWC and SWRC ontologies (some #7 was created with the objective of being capable of net-
of them used by only 1 dataset other than lak ). working the wide range of resources and information held by
libraries and other cultural institutions in German-speaking
The datasets with more than 5 features shared with lak are countries. This dataset uses established vocabularies, such
shown in Table 1. The more expressive result is obtained as FOAF. Dataset #9 describes e-prints and has links cata-
by the rkb-explorer-webconf dataset which shares 31 fea- logued in the DataHub for other bibliographic datasets such
tures with lak. This was the most correlated dataset with as Citeseer, DBLP, ACM and IEEE. Dataset #10 is also
the selected classes and properties of lak. The rkb-explorer- a Linked Data version of publications information of the
webconf is a semantic repository that publishes RDF linked DBLP digital library, similar to sweto-dblp.
data and co-reference information from the RKB Explorer
initiative. This dataset includes information about authors
and publications in several conferences, such as ESWC. Discussion. Based on the top 10 rankings of both ap-
proaches, we identified three main groups of candidate
datasets that were recommended to be interlinked with lak :
4.4 Dataset Interlinking recommendations
Using the score functions, briefly described in Section 3,
we generated recommendations for lak. A preview of the • generic: semanticweb-org, w3c-wordnet, tags2con-
21/02/14
secold acorn-sat
ifpri-linked-open-data-global-hunger-index
rkb-explorer-lisbon
miguiadeviajes-gnoss
sweto-dblp
rkb-explorer-newcastle
rkb-explorer-ibm toxcast-toxrefdb proyectoapadrina
rkb-explorer-acm rkb-explorer-cordis
rkb-explorer-roma twc-healthdata
chronicling-america
osm-semantic-network rkb-explorer-irit
rkb-explorer-italy vivo-indiana-university rkb-explorer-deploy bibsonomy
rkb-explorer-deepblue
rkb-explorer-unlocode libver
vivo-university-of-florida rkb-explorer-courseware datagov-catalog
rkb-explorer-pisa educationalprograms_sisvu southampton-ac-uk-jargon
geospecies
business_terms
rkb-explorer-epsrc
rkb-explorer-ulm hellenic-police
rkb-explorer-ieee swrc:abstract
rkb-explorer-ft rkb-explorer-nsf
dce:creator jita
bibo:authorList
swc:relatedToEvent
productontology
rkb-explorer-era
swc:isPartOf
rkb-explorer-rae2001 rkb-explorer-budapest dce:subject swrc:year
swc:hasPart
rkb-explorer-darmstadt
dce:titleglottolog-langdoc newsweek
oecd-linked-data swrc:month
rkb-explorer-southampton
rkb-explorer-laas open-data-euskadi
rkb-explorer-risks hebis-bibliographic-resources
iso-3166-2-data swc:hasAcronym
world-bank-linked-data temple-ov-thee-lemur-datasets
rkb-explorer-digitaleconomy swrc:InProceedings
hedatuz
jamendo-dbtune
foaf:made
iris2 landscape-portrait
rkb-explorer-kaunas
rkb-explorer-eurecom linked-open-vocabularies-lov
ndaa2011 psh-subject-headings glastonbury-2011
garnicaplywood swc:completeGraph
swrc:booktitle
lak
bizkaisense swrc:affiliation foaf:based_near
public-record-office-victoria-semantic-wiki sudocfr
interactivemaps-gnoss rkb-explorer-webconf
instance-hub-us-civil-servants southampton-ac-uk-pressinfo
rkb-explorer-jisc
hellenic-fire-brigade
rdfohloh foaf:maker
geological-survey-of-austria-thesaurus
nytimes-linked-open-data aksworg greek-legal-entities
rkb-explorer-citeseer
idreffrsouthampton-ac-uk-photos foaf:homepage
southampton-ac-uk-profile
dcs-sheffield swc:hasRelatedDocument
dct:subject krystian-pietruszka
swc:ConferenceEvent
camera-deputati-linked-data
linked-structured-product-labels
fao-geopolitical-ontology
foaf:name
ontos-news-portal
dbtune-john-peel-sessions foaf:member
foaf:mbox_sha1sum
msc foaf:Person euskadi-farmacias
instance-hub-us-congressional-committees
movies-argentina
courts-thesaurus
swrc:Proceedings
debian-package-tracking-system
nobelprizes southampton-ac-uk-apps
event-media open-food-facts rkb-explorer-webscience
foaf:Organization lod2
cablegate
southampton-ac-uk-phonebook
german-labor-law-thesaurus swrc:series
twc-data-gov semanticweb-org
beneficiaries-of-the-european-commission
transparency-linked-data foaf:firstName eu-parliament-media
national-diet-library-authorities
sparql-endpoint-status
instance-hub-people
foaf:lastName
eurostat-rdf southampton-ac-uk-services taxonconcept
kdata national-diet-library-subject-headings
instance-hub-organizations sandrart-net
vivo-weill-cornell-medical-college southampton-ac-uk-org
qualitywebdata-org vivo
instance-hub-us-federal-agencies
arrayexpress_e-mtab-104
vivo-scripps-research-institute vivo-ponce
enipedia
Figure 3: The datasets and their associated lak features.
delicious, geonames-semantic-web, lexvo, nytimes- with smaller popularity and having at least one feature of
linked-open-data, rkb-explorer-wiki lak.
• bibliographic: dcs-sheffiedl, linked-open-camera, The results also indicate that the selection of the feature set
sweto-dblp, rkb-explorer-acm, lcsh, dnb-gemeinsame- is very important because it directly influences the generated
normdatei, rkb-explorer-eprints, rkb-explorer-dblp rankings and can lead to recommendations of datasets which
are more as well as less generic. In our experiments with lak,
• educational area: gnoss. we filtered out some generic features (e.g., owl:sameAs), but
included DC and FOAF elements. Thus, we expected that
both generic and specific datasets from our set of datasets
The top 10 recommendations of the rankings differ in were recommended. As the metadata used to triplify lak
some aspects. Considering the groups identified above, were not using classes and properties specifically related
the Bayesian ranking contains a higher number of generic to the application domain, this characteristic was not ev-
datasets, while the Social Network-based ranking contains idenced in the recommendation results.
a higher number of bibliographic datasets. This probably
happens because Bayesian ranking prioritizes recommenda- 5. CONCLUSIONS
tions for lak of datasets linked from the larger number of This paper presented a detailed analysis, based on Bayesian
other datasets having the larger number of lak features. On classifiers and on Social Network Analysis techniques, to ad-
the other hand, the Social Network-based ranking prioritizes dress the dataset interlinking recommendation problem for
the datasets pointed by the larger number of other datasets lak, using only metadata. Thus, the rank score functions are
Table 2: Top 10 ranked recommendations for lak.
# Bayesian ranking score∗ # SN-based ranking score
1 semanticweb-org -162.025 1 geonames-semantic-web 13.738
2 w3c-wordnet -162.236 2 nytimes-linked-open-data 3.558
3 tags2con-delicious -163.025 3 gnoss 3.051
4 dcs-sheffield -163.025 4 lcsh 3.017
5 linked-open-camera -163.025 5 rkb-explorer-acm 2.430
6 sweto-dblp -163.025 6 rkb-explorer-wiki 2.408
7 geonames-semantic-web -3281.339 7 dnb-gemeinsame-normdatei 2.020
8 lexvo -4107.754 8 lexvo 2.017
9 rkb-explorer-acm -4114.493 9 rkb-explorer-eprints 1.632
10 lcsh -4273.558 10 rkb-explorer-dblp 1.466
∗
Estimated using log2 , c=-170 and considering only lak features shared with at least one dataset.
Casanova, and S. Dietze. Recommending tripleset
interlinking through a social network approach. In
WISE’13, pages 149–161, 2013.
[5] G. R. Lopes, L. A. P. Paes, B. P. Nunes, M. A.
Casanova, and S. Dietze. Comparing recommendation
approaches for dataset interlinking. Technical report,
Department of Informatics, PUC-Rio, 2013.
[6] B. F. Lóscio, M. Batista, and D. Souza. Using
information quality for the identification of relevant
web data sources. In IIWAS’12, pages 36–44, New
York, NY, USA, 2012. ACM.
[7] L. Lü, C.-H. Jin, and T. Zhou. Similarity index based
on local paths for link prediction of complex networks.
Physical Review E, 80(4):046122, 2009.
[8] C. D. Manning and H. Schütze. Foundations of
Statistical Natural Language Processing. MIT Press,
2002.
[9] A. Nikolov and M. d’Aquin. Identifying Relevant
Sources for Data Linking using a Semantic Web Index.
In LDOW’11, 2011.
Figure 4: Preview of the RecLAK recommendation
[10] A. Nikolov, M. d’Aquin, and E. Motta. What Should I
interface.
Link to? Identifying Relevant Sources and Classes for
Data Linking. In JIST’12, pages 284–299. Springer
potentially useful to reduce the cost of dataset interlinking. Berlin Heidelberg, 2012.
For more information, including the full set of data used [11] B. P. Nunes, B. Fetahu, and M. A. Casanova.
in the experiments, graphical visualizations and detailed re- Cite4me: Semantic retrieval and analysis of scientific
sults, we refer to the RecLAK Web application, avaliable at publications. In LAK (Data Challenge), volume 974 of
http://www.inf.puc-rio.br/~grlopes/RecLAK. CEUR Workshop Proceedings. CEUR-WS.org, 2013.
[12] B. P. Nunes, B. Fetahu, S. Dietze, and M. A.
6. ACKNOWLEDGMENTS Casanova. Cite4me: A semantic search and retrieval
web application for scientific publications. In ISWC
This work was partly funded by CNPq, under grants
(Posters & Demos), volume 1035 of CEUR Workshop
160326/2012-5, 301497/2006-0 and 57128/2009-9, and
Proceedings, pages 25–28. CEUR-WS.org, 2013.
by FAPERJ, under grants E-26/170028/2008 and E-
26/103.070/2011. [13] H. R. d. Oliveira, A. T. Tavares, and B. F. Lóscio.
Feedback-based data set recommendation for building
linked data applications. In I-SEMANTICS’12, pages
7. REFERENCES 49–55, 2012.
[1] T. Berners-Lee. Linked Data. In Design Issues. W3C, [14] D. Taibi and S. Dietze. Fostering analytics on learning
July 2006. analytics research: the lak dataset. In LAK (Data
[2] K. A. Kuznetsov. Scientific data integration system in Challenge), volume 974 of CEUR Workshop
the linked open data space. Programming and Proceedings. CEUR-WS.org, 2013.
Computer Software, 39(1):43–48, Jan. 2013. [15] A. Wagner, P. Haase, A. Rettinger, and H. Lamm.
[3] L. A. P. P. Leme, G. R. Lopes, B. P. Nunes, M. A. Discovering related data sources in data-portals. In
Casanova, and S. Dietze. Identifying candidate SemStats workshop, ISWC’13, 2013.
datasets for data interlinking. In ICWE’13, pages
354–366, 2013.
[4] G. R. Lopes, L. A. P. P. Leme, B. P. Nunes, M. A.