=Paper=
{{Paper
|id=Vol-1184/paper8
|storemode=property
|title=Will Linked Data Benefit from Inverse Link Traversal?
|pdfUrl=https://ceur-ws.org/Vol-1184/ldow2014_paper_08.pdf
|volume=Vol-1184
|dblpUrl=https://dblp.org/rec/conf/www/ScheglmannS14
}}
==Will Linked Data Benefit from Inverse Link Traversal?==
https://ceur-ws.org/Vol-1184/ldow2014_paper_08.pdf
Will Linked Data Benefit from Inverse Link Traversal?
Position Paper
Stefan Scheglmann Ansgar Scherp
University of Koblenz-Landau, Germany Kiel University, Germany
Institute for Web Science and Techologies Leibniz Information Center for Economics
schegi@uni-koblenz.de asc@informatik.uni-kiel.de
ABSTRACT proofs of concepts, we now more and more see the devel-
Query execution using link-traversal is a promising approach opment of common frameworks and standardization efforts
for retrieving and accessing data on the web. However, this around the four LOD principles. One example is the cur-
approach finds its limitation when it comes to query patterns rent W3C working draft on a Linked Data Platform (LDP)2,3 ,
such as ?s rdf:type ex:Employee, where one does not know which aims at providing further clarifications and extensions
the subject URI. Such queries are quite useful for different to the four Linked Data Principles defined by Tim Berners-
application needs. In this paper, we conduct an empirical Lee. It describes a generic infrastructure for providing LOD
analysis on the use of such patterns in SPARQL query logs. resources as well as modifying them. The principle idea is
We present different solution approaches to extend the cur- to offer so-called LDP resources (LDPR) that can be derefer-
rent Linked Open Data principles with the ability for inverse enced in order to obtain information about it (read), modify
link traversal. We discuss the advantages and disadvantages or delete the resource, or create a new resource. In addition
of the different approaches. to the notion of LDPR, the draft also suggests the concept of
LDP collections (LDPC). A LDPC represents a set of “same-
subject, same-predicate triples”, which can be accessed and
Categories and Subject Descriptors modified through a common URI. Another example of a stan-
H.1 [Information Systems Applications]: Models and Prin- dardization effort is the SPARQL Graph Protocol4 , a W3C rec-
ciples; H.4 [Information Systems Applications]: Miscella- ommendation that operates on the level of graphs as entities
neous for creation, modification, and deletion. Both specifications
overlap in certain points and seek to align their efforts as
much as possible.5
General Terms We very much appreciate these efforts as they allow for de-
Design, Measurement veloping more robust and mature LOD services. However,
we believe that still improvements are needed regarding an
Keywords efficient access and retrieval of linked data. Searching on
linked data can in principle be categorized into two different
Querying of Linked Open Data; Link Traversal approaches: a) Crawling all the data in a first step and sub-
sequently indexing it to make it available to the users [6, 12,
1. INTRODUCTION 19, 15, 5, 18, 9] and b) On-the-fly retrieval of the data accord-
The four principles of Linked Open Data (LOD) as pub- ing to the follow-your-nose-principle [11, 10]. While the first
lished in 20061 have gained widespread adoption in research, approach adopts the typical search scenario as it is known
industries, as well as governmental and non-profit organi- from the web, the second approach comes more natural with
zations. Today, we find various approaches for publish- the LOD approach. This link traversal based query execu-
ing linked data such as dedicated data stores like DBpedia, tion seems one of the most promising approaches to execute
SPARQL engines with a SPARQL endpoint, plain RDF files, queries over the Web of Linked Data. No fixed, predefined
as well as data embedded into websites such as RDFa. After set of relevant data sources has to be defined and relevant
the “early years” of individually implemented solutions and sources can be discovered during query execution. Hartig et
al. [11] defined the semantics of link-traversal query execu-
1
http://www.w3.org/DesignIssues/LinkedData.html last tion on the web of data. However, it has its limitations. In
visit 16th February, 2014
2
http://www.w3.org/TR/2013/WD-ldp-20130730/ last visit
16th February, 2014
3
https://dvcs.w3.org/hg/ldpwg/raw-file/default/
ldp-bp/ldp-bp.html last visit 16th February, 2014
4
http://www.w3.org/TR/sparql11-http-rdf-update/ last
visit 16th February, 2014
5
A detailed discussion on the overlaps and resulting con-
flicts can be found here: http://www.w3.org/2012/ldp/
wiki/Main_Page#Linked_Data_Platform_.28LDP.29_vs_
Copyright is held by the author/owner(s). SPARQL_Graph_Store_HTTP_Protocol_.28GSP.29 last visit
LDOW2014 April 8, 2014, Seoul, Korea. 16th February, 2014
order to illustrate this, let us consider a simple example taken where link traversal based query execution is not sufficient.
from Hartig and Freytag [11] as shown in Listing 1 below. But this might not always be possible. According to the state
of the LOD cloud7 in 2011, only 68% of the data sources in the
LOD cloud provide SPARQL endpoint to allow expressive
Listing 1: Example query on LOD (from [11]).
1 SELECT ?p ?l WHERE { queries to be asked against the datasets. The remaining 31%
2 ?p. are only accessible by link traversal queries.
3 ?p ?pr. But also the stability of provided SPARQL endpoints is of
4 ?pr ?l. interest. Buil-Aranda et al. [4] studied the current state of
5 } available public SPARQL endpoints. They conducted vari-
The query specifies the information need to obtain a list of ous experiments to test discoverability, interoperability, per-
project names that friends of work on. This formance, and availability. According to them availability
query can be executed over LOD as shown by the authors [11]. of SPARQL endpoints is still an issue. In their experiments,
Still, executing queries on LOD has significant limitations they monitored 427 public SPARQL endpoints which are reg-
when it comes even to slighty different information needs. istered at DataHub8 over a 27-month long experiment. Re-
For example, we might not know a-priori the existence of garding the availability, they conclude that only around 32%
Bob or even if we know about him as person, we might not of the endpoints reach an uptime of 99−100%, 59% an uptime
be aware of being the URI representing over 75%, and 29.3% are available less than 5% of the time.
him. Thus, we like to execute a query on LOD as shown in In order to answer Question 2, we rely on SPARQL query
Listing 2 where we do not know the subject URI of the first logs. Different analyses about the current usage of SPARQL
query pattern. Such a query is useful, e. g., on data providers already provide a well substantiated line of argumentation.
such as the fictitious media company Big Lynx described in For instance, Gallego et al. [2] analyzed the USEWOD2011
the Linked Data book by Heath and Bizer6 to first determine dataset. Their results show that more than 90% of the queries
which employees are working with the company, before finding consist of less than 3 query patterns. SPARQL query patterns
out further information about each person. with the subject unbound while given predicate and object
corresponds to our rdf:type pattern. They are the third most
used type of patterns, with 7% in DBpedia and 46% in Se-
Listing 2: Query requiring inverse traversal. mantic Web Dog Food (SWDF)9 conference metadata. Möller
1 SELECT ?s WHERE
et al. [14] applied similar analysis on a different dataset. They
2 ?s
3 . conducted an analysis on query logs of SWDF, DBpedia, DB-
4 } tune, and RKBExplorer (RKB). In their analysis they came
to comparable results, 90% consist of less than three triple
However, this and similar queries cannot be executed on patterns. The query pattern with unbound subject is used in
the LOD today as it requires an inverse link traversal from the 43% of the SWDF queries, 68% of the DBTune queries, 68%
Big Lynx specific vocabulary defining the concept Employee of the RKB queries, and 50% of the DBpedia queries.
to the resource URIs that are typed with this concept. Please Nethertheless, we conduced our own analysis, in order
note that our discussion of inverse traversal mainly addresses to show the frequent use of rdf:type in SPARQL query pat-
the rdf:type triples. However, the ideas can be applied in terns. We extracted SELECT queries taken from the USE-
principle to triples with any kind of properties. We refer to WOD201310 data challenge files. The USEWOD2013 queries
this at some parts of the paper. where taken from Apache CLF11 logs of four different linked
Below, we first further illustrate and motivate the need open data sources, Linked Geo Data (LGD)12 , DBpedia13 ,
for an local inverse link traversal of rdf:type predicates on bio2RDF14 , and Semantic Web Dog Food (SWDF) conference
LOD. Subsequently, we conduct an empirical analysis on the metadata. Table 1 shows the results of our analysis.
use of SPARQL queries involving query patterns such as the We could show that around 11% of all SELECT queries
one shown in Listing 2. We present possible approaches to from the aforementioned logs contain at least one pattern
extend the current LOD principles with the ability for inverse with rdf:type as predicate. This ratio ranges from less than
link traversal and discuss its advantages and disadvantages. 1% in the bio2rdf queries up to 14.1% in the DBpedia queries.
Finally, we present related work before we conclude. Please note that there is also the possibility that a triple pattern
like ?pr rdf:type ex:ResearchProject is contained in queries
2. QUERIES ON THE SEMANTIC WEB like that one in Listing 1. In such cases, the query does not
In order to motivate the need for an inverse link-traversal 7
http://lod-cloud.net/state/ last visit 16th February,
for querying LOD, we first look into further detail how queries 2014
on the Semantic Web are executed. To this end, we consider 8
http://datahub.io/en/dataset?res_format=api\
and discuss the following two questions: %2Fsparql last visit 16th February, 2014
(1) Why is it not sufficient to fall back to other query paradigms 9
http://data.semanticweb.org last visit 16th February,
like SPARQL in cases where we need to resolve query pat- 2014
10
terns such as the example in Listing 2. http://data.semanticweb.org/usewod/2013/ last visit
(2) Is this pattern important enough and would the local exe- 16th February, 2014
11
cution generate enough benefit to justify such an intervention Common Log Format, an informal standardhttp://en.
into existing standards? wikipedia.org/wiki/Common_Log_Format last visit 16th
February, 2014
First, we address Question 1: As already mentioned, one 12
http://linkedgeodata.org/About last visit 16th February,
might argue that one should fall back to SPARQL in cases 2014
13
6
http://linkeddatabook.com/editions/1.0/ last visit http://DBpedia.org/About last visit 16th February, 2014
14
16th February, 2014 http://bio2rdf.org/ last visit 16th February, 2014
DataSet #SELECT #Queries w. Ratio For inverse rdf:type traversal like in Listing 2 such an URI with
DataSet Queries rdf:type % an embedded query might look like the examples displayed
LGD 1,639,889 139,788 8.52 in Listing 4. Query (1) introduces “instanceOf” as keyword
DBpedia 27,089,369 3,819,413 14.10 to indicate inverse of rdf:type. This would allow to retrieve
SWDF 11,157,747 556,783 4.99 all instances of a specific RDF type. Query (2) extends this
bio2rdf 179,370 1 0.00 to “subjectOf” and allows to specify with another parameter
total 40,066,375 4,515,985 11.27 the property URI like here the rdf:type. This allows to embed
queries for arbitrary inverse properties.
Table 1: SPARQL query statistics on the USEWOD2013
dataset
Listing 4: Embedded Queries
(1) http :// ex.com/ instanceOf?
necessarily need the possibility to inverse traversal of rdf:type
links. However, investigating these cases was beyond scope (2) http :// ex.com/ subjectOf?
of our analysis. p=&
o=
3. POSSIBLE SOLUTION APPROACHES In order to address the problem that a data provider hosts a
In Linked Data, each resource has an explicit URI which very large number of resources of specific RDF type, the data
can be resolved in order to retrieve additional information can be retrieved in a paginated manner. For example, the sim-
about this entity. However, how can one do this for RDF ple HTTP GET query can be extended by determining a limit
types, i. e., how to dereference RDF type URIs? Here, two and an offset for the data, e. g., http://ex.com/instancesOf?
principal challenges have to be addressed: http://.../BigLynx/Employee&offset=100&limit=40.
This retrieves forty instances starting at an offset of 100.
• RDF types used in linked data sources are often not de- The queries above could be provided as new features simi-
fined by the source itself, e. g., foaf:Person. Thus, the lar to the URI lookup endpoint feature of VoiD.16 This feature
provider hosting the vocabulary does not know any- allows to determine a dedicated URI that can be used for
thing about the data sources that are using the vocabu- search by appending the keywords to this URI.
lary to describe the resources.
• In addition, resolving RDF types such as foaf:Person in Approach 2: Dedicated Schema URI. Instead of provid-
order to retrieve its instances can potentially lead to a ing a query mechanism, information about rdf:type entities
very large amount of results. In an extreme case, i. .e, could also be made accessible by a specific URI per type. De-
when we aim at resolving the RDF type rdf:Resource, pending on the number of types in a dataset, this could lead
we end up retrieving all instances. to a potentially large number of entity-URIs. An alternative
solution is that linked data providers could provide a special
In order to address the challenges, one cannot operate on a kind of RDF schema document. This schema document has to
global level, i. e., on the entire Web of Linked Data. Rather, be made available under an generally accepted schema URI,
we aim to provide resources of RDF types only on a local e. g., http://example.org/.well-known/schema/instance-types.17
level, i. e., resources hosted by data sources that are run by a A schema document consists of triples in the form rdf:type , thereby the entities-URI provides
Thus, when resolving a RDF type such as foaf:Person, a data a direct lookup mechanism to find instances of the specific
provider operating a specific pay-level domain may return type. For example, assuming two entities of foaf:Person on
a reference containing information about all instances it de- a server, namely http://bob.name and http://tim.name, its en-
fines using this type, e. g., by the semantic pingback mecha- tities document would consist of two triples, cf. Listing 5.
nism [17]. In summary, we can come up with the following If this list is very long, i. e., there exist a large number of in-
solution approaches based on common REST practices to al- stances, a pagination approach can be implemented by con-
low inverse link traversal on Linked Data: necting multiple RDF schema documents via rdf:List.
Approach 1: Simple URI Queries. It is possible that linked
data providers extend their servers with mechanisms that al- Listing 5: Entities document for foaf:Person
rdf:type .
low for HTTP GET requests with embedded queries. This is rdf:type .
inspired by the W3C standardization of Media Fragments15 ,
i. e., the definition of a unique URI scheme to access frag- A globally agreed URI schema like http://example.org/.well-
ments of media assets such as images and videos on the web. known/schema allows to instantly access any kind of relevant
A media fragment URI specification supports a lightweight schema information for the entities in a data source such as
mechanism to embed queries in such URIs. According to the instances of specific types. However, it is debatable whether
standard every URI consists of four parts: standardization efforts should agree on generally accepted
URIs for storing schema information. Thus, another way
Listing 3: Media Fragment URI Schema 16
1 : http://www.w3.org/TR/2011/NOTE-void-20110303/
2 [ ? ] [ # ] #lookup last visit 16th February, 2014
17
Please note the use of the path .well-known/ for modeling a
15
http://www.w3.org/TR/media-frags/ “well-known location” for common data like our schema data
#standardisation-URI-queries last visit 16th Febru- as proposed in RFC 5785 available from: https://tools.
ary, 2014 ietf.org/html/rfc5785
would be to add a link to an instances list to a VoID metadata 4. RELATED WORK
description of the datasets. We find a plethora of different approaches for searching
Linked Open Data (and the Semantic Web in general) such
Implementation Issues. as Swoogle [6], SemSearch [12], Sig.ma [19], Sindice [15], Fal-
Finally, we have to think about how the response to a query cons [5], Hermes [18], and LODatio [9].20 Despite particular
(Approach 1) or the type-entity documents (Approach 2) can features and differences these systems have, they all have in
be generated. In general, Linked Data can be provided in common that they index a crawled set of semantic data. As
different means. One can distinguish three different ways of such, they require the semantic data readily at hand for search
publishing and accessing LOD on the web: in order to provide answers to the users’ queries. LODatio
(a) Linked data sources using a dedicated RDF infrastructure, does not store the instance data itself but solely relies on a
like a triple store in the back-end. These sources may pro- schema-level index. This schema-level index allows to find
vide a SPARQL endpoint to enable for complex queries. An sources of information on the web of data without keeping
example of such a source is DBpedia. the original data. However, it still requires an initial crawl of
(b) Linked data may also be provided by sources as a set LOD to compute the schema-level index.
of different RDF files or data dumps. An example for this Among the different approaches for semantic search, there
could be a web server just providing a couple of FOAF18 are also some that exploit information from existing query
documents, in addition to personal information published logs. Examples are Adeyanju et al. [1] who use query logs
on hosted websites. for query term recommendations such as generalizations and
(c) Linked data could even be embedded into existing (X)HTML specializations. Meij et al. [13] align query logs with concepts
documents of a website using RDF in attributes (RDFa)19 . from DBpedia and conduct recommendations based on the
For sources of type (a) the response to an inverse link traver- number of concepts linking to or from these concepts. Over-
sal could just be generated by executing a query correspond- all, query logs provide useful information about the actual
ing to the request directly on the triple store. Sources of type information needs a LOD client has. However, they are not
(b) and (c) need a more sophisticated mechanism. Servers used by LOD providers to compute some a-priori available
providing linked data in that way have to somehow collect indices that ease the consumption of their data. Finally, we
the information first. If the linked data is provided in multi- find tools that solely rely on exploring the data in direct com-
ple RDF files, e.g. like on a FOAF server, we might extend the munication with the LOD providers. A well known example
server by an indexing mechanism, which provides all RDF is the Tabulator system by Tim Berners-Lee [3] that allows be-
files and generates the response set. In order to also support sides viewing and browsing also editing and thus updating
embedded RDF, this indexing mechanism has to be extended the data. Also Marbles21 allows for browsing the LOD graph
so that it can provide arbitrary files that allow to embed RDF by following the outgoing edges. Given a user provided URI,
and identify and extract the desired information from those it retrieves all information about it by dereferencing it. In ad-
files. dition, Marbles queries search engines such as Sindice and
Should an indexing service generate the pages dynami- Falcons and follows RDF predicates like owl:sameAs and
cally or should we go for static index pages? The former rdfs:seeAlso to gain further information about a resource.
would be space saving and more robust regarding changes While the data is retrieved live, Marbles can be considered
in the data but has several disadvantages in terms of compu- as hybrid solution as it not only makes use of other seman-
tational power and response times. The latter is clearly more tic search engines but also makes the graph data persistent
space consuming. Depending on the data, the worse case across user sessions to allow for a more efficient access to
leads to redundancy in the URIs that is linear to the num- the data. However, it does not allow to browse the data
ber of used types. It is also less robust regarding changes in providers, e. g., by inverse type links or some simple look-
the data. There might be indices affected by the change in up features to find the instances of particular RDF types on
the data and those have to be updated. But this method has a LOD data source. A comprehensive overview of further
clear advantages regarding performance and response times, LOD components and clients can be found online and in-
because all required answer sets already exist. The concrete cludes Linked Data browsers, crawlers, data extractors, and
decision on how such an indexing service is finally imple- mashup frameworks.22 .
mented depends on the concrete use case and the data. A Hartig et al. [11] formalized traversal-based query exe-
static indexing brings some advantages if space is not critical cution on LOD, provide semantics for queries and analyse
but computational power, or the data is static and changes characteristics of such queries. Examples of graph traversal
infrequent, or the amount of entities per type is reasonable, or languages for RDF data are nSPARQL [16], a language with
there are strong requirements regarding the response times. focus on navigation through RDF data. With NautiLOD [7],
In other cases, e.g., for frequently changing data or datasets Fionda et al. introduced a declarative language which al-
with very large amounts of entities per type, dynamic index- lows to specify portions of the web of data, defines routes
ing might be more convenient. We could also think of adap- and instructs agents to perform actions. They also introduce
tive indexing strategies, e.g. computing only those indexes
dynamically which lead to a high amount of redundancy and 20
For a comprehensive overview, please refer to: http:
precompute indexes for frequently requested types. //www.w3.org/wiki/TaskForces/CommunityProjects/
LinkingOpenData/SemanticWebSearchEngines last visit
16th February, 2014
21
http://mes.github.io/marbles/ last visit 16th February,
18
The friend of a friend ontology http://www.foaf-project. 2014
org/ last visit 16th February, 2014 22
http://www.w3.org/wiki/TaskForces/
19
The RDFa Primer http://www.w3.org/TR/ CommunityProjects/LinkingOpenData/SemWebClients
xhtml-rdfa-primer/ last visit 16th February, 2014 last visit 16th February, 2014
swget, a tool that allows for the use of NautiLOD on the [4] C. Buil-Aranda, A. Hogan, J. Umbrich, and P.-Y.
web of data. GuLP [8] is a query execution language and Vandenbusshe. Sparql web-querying infrastructure:
framework based on the link-traversal paradigm, which can Ready for action? In Proceedings of the 12th International
declaratively include preferential attachment into its queries Semantic Web Conference, Sydney, Australia, 2013.
to order the answer in terms of node/edge attributes. With [5] G. Cheng, W. Ge, and Y. Qu. Falcons: searching and
SQUIN [10], Hartig presented a query execution framework browsing entities on the semantic web. In WWW, pages
that integrates the traversal of data links into the result con- 1101–1102. ACM, 2008.
struction process. All these approaches would clearly benefit [6] L. Ding, T. W. Finin, A. Joshi, R. Pan, R. S. Cost, Y. Peng,
from our proposed extension. P. Reddivari, V. Doshi, and J. Sachs. Swoogle: a search
The VoID Vocabulary23 allows for describing Linked Datasets and metadata engine for the semantic web. In CIKM.
in terms of providing metadata information such as a SPARQL ACM, 2004.
endpoint location, example resources, and the number of [7] V. Fionda, C. Gutierrez, and G. PirrŸ. Semantic
triples, entities, classes, and properties in the data set. How- navigation on the web of data: specification of routes,
ever, this does not help in identifying resources of a specific web fragments and actions. In A. Mille, F. L. Gandon,
type. Most relevant to our work is the VoiD feature to denote J. Misselis, M. Rabinovich, and S. Staab, editors, WWW,
root resources (named using the void:rootResource property). pages 281–290. ACM, 2012.
Here, it is assumed that the “entire dataset can be crawled by [8] V. Fionda and G. Pirrò. Querying graphs with
resolving the root resource(s) and recursively following links preferences. In CIKM, pages 929–938, 2013.
to other URIs in the retrieved RDF responses”. Thus, in prin-
[9] T. Gottron, A. Scherp, B. Krayer, and A. Peters. Lodatio:
ciple it is possible to list all relevant entities here. However,
using a schema-level index to support users infinding
it cannot be stated that specific resources are of a particular
relevant sources of linked data. In K-CAP, pages
RDF type. In addition, finding the relevant resources requires
105–108. ACM, 2013.
that the data has to be crawled first.
[10] O. Hartig. Squin: a traversal based query execution
system for the web of linked data. In K. A. Ross,
5. CONCLUSIONS D. Srivastava, and D. Papadias, editors, SIGMOD
In our opinion traversal-based query execution is one of the Conference, pages 1081–1084. ACM, 2013.
most promising approaches to execute queries over the Web [11] O. Hartig and J.-C. Freytag. Foundations of traversal
of Linked Data. The enormous freedom of this approach, based query execution over linked data. In HT, pages
its flexibility and reliability, the independence from central- 43–52. ACM, 2012.
ized indexes, the on-the-fly discovery of new sources and [12] Y. Lei, V. S. Uren, and E. Motta. Semsearch: A search
its integration with common Web protocols, has the poten- engine for the semantic web. In EKAW, volume 4248 of
tial to make this paradigm the preferred mechanism to ex- Lecture Notes in Computer Science, pages 238–245.
ecute queries on Linked Data. From our point of view this Springer, 2006.
paradigm has still some weaknesses. The analysis of SPARQL [13] E. Meij, M. Bron, L. Hollink, B. Huurnink, and
query logs has shown that inverse traversal of links in RDF M. de Rijke. Learning semantic query suggestions. In
data is crucial in order to make full use of the potential of ISWC, pages 424–440. Springer, 2009.
Linked Data. From our point of view, the possibility to in- [14] K. Möller, M. Hausenblas, R. Cyganiak, and G. A.
verse traversal of rdf:type links would imply a tremendous Grimnes. Learning from linked open data usage:
extension to the possibilities of traversal-based query execu- Patterns & metrics. In Proceedings of the WebSci10:
tion. We propose a set of ideas to ongoing standardization Extending the Frontiers of Society On-Line, 2010.
efforts of linked data and hope that these ideas will lead to [15] E. Oren, R. Delbru, M. Catasta, R. Cyganiak,
a discussion in the community. Part of our future work is H. Stenzhorn, and G. Tummarello. Sindice.com: a
to conduct some experimental research that actually investi- document-oriented lookup index for open linked data.
gates the merits and drawbacks of the approaches outlined IJMSO, 3(1):37–52, 2008.
in the paper. [16] J. Pérez, M. Arenas, and C. Gutierrez. nSPARQL: A
navigational language for RDF. J. Web Sem.,
6. REFERENCES 8(4):255–270, 2010.
[17] S. Tramp, P. Frischmuth, T. Ermilov, and S. Auer.
[1] I. A. Adeyanju, D. Song, M.-D. Albakour,
Weaving a Social Data Web with Semantic Pingback. In
U. Kruschwitz, A. De Roeck, and M. Fasli. Adaptation
Proceedings of the EKAW 2010, pages 135–149, Berlin /
of the concept hierarchy model with search logs for
Heidelberg, October 2010. Springer.
query recommendation on intranets. In SIGIR, pages
5–14. ACM, 2012. [18] T. Tran, H. Wang, and P. Haase. Hermes: Data web
search on a pay-as-you-go integration infrastructure. J.
[2] M. Arias, J. D. Fernández, M. A. Martínez-Prieto, and
Web Sem., 7(3):189–203, 2009.
P. de la Fuente. An Empirical Study of Real-World
SPARQL Queries. CoRR, abs/1103.5043, 2011. [19] G. Tummarello, R. Cyganiak, M. Catasta, S. Danielczyk,
R. Delbru, and S. Decker. Sig.ma: Live views on the
[3] T. Berners-Lee, J. Hollenbach, K. Lu, J. Presbrey,
web of data. J. Web Sem., 8(4):355–364, 2010.
E. Prud’hommeaux, and M. M. C. Schraefel. Tabulator
redux: Browsing and writing linked data. In LDOW,
volume 369, 2008.
23
http://www.w3.org/TR/2011/NOTE-void-20110303/ last
visit 16th February, 2014