Download Integrated Access to Cultural Heritage

Integrated Access to Cultural Heritage Resources through
Representation and Alignment of Controlled Vocabularies
Antoine Isaac
Antoine Isaac (http://www.few.vu.nl/~aisaac/) works as a postdoc at the Vrije
Universiteit in Amsterdam and the Koninklijke Bibliotheek in the Hague, in the context
of the STITCH and TELplus projects. His research interests include different aspects
of the use of Semantic Web languages and technologies within the Cultural Heritage
field, focusing on the representation and the interoperability of collections and their
vocabularies.
Address: Antoine Isaac, Vrije Universiteit Amsterdam, Department of Computer
Science, De Boelelaan 1081a, 1081 HV Amsterdam, The Netherlands.
Email: [email protected]
Abstract:
Purpose of this paper
To show how Semantic Web techniques can help addressing semantic interoperability issues
in the broad cultural heritage domain allowing users an integrated and seamless access to
heterogeneous collections.
Design/methodology/approach
This paper presents the heterogeneity problems to be solved. It introduces Semantic Web
techniques that can help solving them, focusing on the representation of controlled
vocabularies and their semantic alignment. It gives pointers to some previous projects and
experiments that have tried to address the problems discussed.
Findings
Semantic Web research provides practical technical and methodological approaches to
tackle the different issues. Two contributions of interest are the Simple Knowledge
Organization System (SKOS) model and automatic vocabulary alignment methods and tools.
These contributions were demonstrated to be usable for enabling semantic search and
navigation across collections.
Originality/value
This article provides a general and practical introduction to relevant Semantic Web
techniques. It is of specific value for the practitioners in the Cultural Heritage and Digital
Libraries domain who are interested in applying these methods in practice.
Research limitations/implications
Our research aims at choreographing different representation and alignment methods for
solving interoperability problems in the context of controlled subject vocabularies. Given the
variety and technical richness of current research in the Semantic Web field, it is impossible
to provide an in-depth account or an exhaustive list of references. Every aspect is however
given one or several pointers for further reading.
Keywords Semantic Web, integrated access, thesaurus alignment, SKOS, semantic
interoperability
1. Introduction: the Semantic Interoperability Problem
In the digital age, cultural heritage (CH) institutions have the opportunity, and face
the challenge, to use the World Wide Web to make accessible the digital artefacts of
their collections, together with their metadata. Web-based access to digitized images
and their descriptions, at anytime, from anywhere, lowers the barriers for access to
information resources. Once there is digital access to the content of museums,
libraries and archives, there is also the tremendous opportunity to merge collections
from different locations into virtual, federated institutions, thus increasing access
across collections and institutional boundaries.
In stark contrast to the vast amount of existing digital resources on the World Wide
Web, cultural heritage assets from libraries, museums, and archives are very well
described. Over many generations, librarians, curators, archivists have developed
knowledge organization systems (KOSs), among which controlled vocabularies such
as thesauri, classification schemes and ontologies, to organize and manage their
collections. The organization and access to cultural heritage along human capacity to
deal with information and knowledge is a valuable human achievement in itself. It
helps us to grasp our past and present, and this understanding must be exploited to
facilitate its access at a grander scale.
The move toward cross-institutional CH portals is well under way as, for instance,
The European Library1 and the Memory of the Netherlands2 testify. In this paper, we
describe how CH expertise can be combined with knowledge and technology from
the Semantic Web (SW) community to deliver portals that provide a seamless and
unified access to different collections via semantic search and navigation.
Fig. 1 illustrates the problem that needs to be solved in a networked environment.
Consider two collections, each of which is indexed by its dedicated knowledge
organization system. Instead of using one single conceptual vocabulary for querying
or browsing the objects of both collections simultaneously, users are expected and
required to use the terminology of the first KOS to identify objects of the first
collection, and the second KOS to identify those of the second collection.
Figure 1 Semantic heterogeneity hampers collection access
1
2
See http://www.theeuropeanlibrary.org
See http://www.geheugenvannederland.nl
We say that these two knowledge organisation systems are not interoperable at the
semantic level. In the given example, when searching for objects showing a
“Madonna” one will only retrieve objects that were indexed using this specific subject
description (the statue in the upper right), and will not find the manuscript illumination
(in the lower right) that was indexed as “Virgin Mary”, which is clearly a conceptually
similar subject description, but stems from another controlled vocabulary.
Not taking care of the semantic heterogeneity of their respective KOSs when merging
collections clearly hampers the ease of accessibility. The burden of search is indeed
transferred to users who then need to perform two well-formulated queries (using the
respective correct terminology) to obtain the desired objects from the two collections.
Two heterogeneity problems must be solved to enhance the interoperability of
controlled vocabularies and, hence, of the systems and collections that use them:
-
Representation heterogeneity : vocabularies often come in different formats:
some will be encoded in XML while others will come as plain text. Beyond, the
models guiding their design might not be directly compatible, for instance,
because they mirror different general information needs (e.g., thesauri contain
“terms” while classification schemes contain “classes”), and different KOS
might have different kinds of notes and labels attached to conceptual entities.
-
Conceptual heterogeneity : any two vocabularies will usually contain concepts
that have identical or similar meanings but different labels or names (e.g., like
“Virgin Mary” and “Madonna”). Also, there will be concepts that are more
general than others (e.g., like “Mother” and “Virgin Mary”). Such similarity and
subsumption links have to be determined and exploited so that an integrated
system can provide users with seamless access to joint content described by
several vocabularies.
In this paper we show how these two problems can be addressed using techniques
that are currently being investigated in the Semantic Web research domain. In
section 2 we describe the basic elements of the Semantic Web infrastructure, and
illustrate how the Simple Knowledge Organization Systems (SKOS) standard model
can be used to represent different vocabularies (KOSs) homogeneously. In section 3,
we show how the representation of the different vocabularies, then commonly
represented in the SKOS format, can be semantically aligned to enable a semantic
integration of different collections. Finally, in section 4, we demonstrate how we
solved a real-life problem with a combination of SW techniques, and briefly describe
the resulting prototype.
2. Semantic Web techniques and controlled vocabulary representation
The Semantic Web (Berners-Lee, Hendler & Lassila, 2001) is a proposed extension
of the existing Web, where information found on the web is augmented with machineaccessible knowledge3. The basic building blocks of the Semantic Web, as
3
The following is a simplified introduction to the Semantic Web. For further detail, the reader is
encouraged to read the Semantic Web Primer (Antoniou & Harmelen, 2004).
introduced by the Resource Description Format (RDF)4, are resources which denote
any element that can be identified on (or even outside) the Web. These resources
are described by three-part statements that link them together. Each statement has
a subject resource which is linked to an object resource via a property resource.
Together, several such triplets form a graph, such as the one represented in Fig. 2 5.
These graphs can contain:
-
-
factual knowledge: the third paragraph of the described document is about
“Amsterdam”; the type of the described document is “Article”; and the selected
paragraph “par3” is part of a (larger) file called “file1”.
ontological knowledge: the Semantic Web is concerned about the way
resources can be grouped in conceptual classes. These classes are
introduced in ontologies that contain formally expressed knowledge about
them. Here, “Article” is a class more specific than (or a subclass of)
“Document”.
Figure 2 A Semantic Web RDF graph.6
The information contained in ontologies is important, since it provides material for
automated reasoning on the resources which populate the classes. For example,
from the information found in Figure 2 for “file1”, “Article” and “Document”, an
automated reasoning engine can infer that “file1” is also an instance of the
“Document“ class, which will yield more answers for queries containing “Document”.
It should be noted that the RDF framework is designed to allow different sources of
knowledge to co-exist with each other, inhabiting the same space. This means that
Semantic Web data can merge and operate with resources coming from different
information spaces. In our example, the objects and links in figure 2 come from
4
See http://www.w3.org/RDF.
Figure 2 is an abstract representation of an RDF graph. Such a graph will be usually serialized in the
form of an XML file, according to the RDF/XML syntax specified by the World Wide Web Consortium
(W3C).
5
6 Nodes in the graph are RDF resources; labelled edges represent assertions of a property between the linked elements. “rdf:”
namespace stands for “http://www.w3.org/1999/02/22-rdf-syntax-ns#”, “rdfs:” for “http://www.w3.org/2000/01/rdf-schema#”,
“myVoc1:” for “http://example.org/voc1#”, and “myVoc2:” for “http://example.org/voc2#”.
different namespaces, either user-defined (myVoc1:, myVoc2:) or predefined
(rdf:). The resource “Amsterdam” in myVoc2: may indeed refer to the capital of
The Netherlands (as the RDF graphs in which it occurs would show), while some
other resource with the same name, but from a different vocabulary space may refer
to a city in the state of New York, USA. In any case, both resources stem from
different name spaces and can both inhabit different contexts further defining and
constraining their intended meaning.
RDF “triples” are the basic building blocks for translating KOS into a homogeneous
format. Also, in order to mirror a KOS’ modelling elements (e.g., the “broader than”,
or “narrower than” relation types of thesauri), additional constructs are necessary.
RDF-Schema (Brickley & Guha, 2004), in short RDF-S, is a simple representation
language that allows users to define their models, introducing different types for RDF
resources and links. One can also express, for instance, that the source and target of
a relation are of a specific type, e.g., that the relation type “has painted” requires a
subject of type “painter” (or “artist”), and an object of type “painting” (or “drawing”).
The current standard web ontology language is called OWL (McGuinness &
Harmelen, 2004). This formal language is more expressive than RDF-S, allowing
users to define a variety of different properties of classes and relations between
them. A more detailed discussion of OWL is beyond the scope of this paper.
To support experts in converting their KOSs into the RDF-based formats, but also to
facilitate the future exchange of such formats, the World Wide Web Consortium
(W3C) has initiated the development of SKOS7, a standard model that allows CH
practitioners (and other terminologists) to homogenously represent the basic features
of knowledge organization systems. SKOS introduces a set of constructs for RDF,
which mainly allow for the description of concepts and concept schemes (Miles &
Brickley, 2005).
Concept description
SKOS has chosen a concept-based approach for the representation of controlled
vocabularies. As opposed to a term-based approach, where terms from natural
language are the first-order elements of a KOS, SKOS describes abstract concepts
that may have a different materialization in language (lexicalizations). SKOS
introduces a special construct skos:Concept8 to properly characterize the (web)
resources that denote such KOS elements. To further specify these conceptual
resources, SKOS features:
-
7
Labelling properties, e.g. skos:prefLabel and skos:altLabel, to link a
concept to the terms that represent it in language. The prefLabel value shall
be a non-ambiguous term that uniquely identifies the concept, and can be
used as a descriptor in an indexing system. The term altLabel is used to
introduce alternative entries – synonyms, abbreviations etc. SKOS allows
concepts to be linked to prefLabels and altLabels in different languages.
SKOS concepts can thus be used seamlessly in multilingual environments.
SKOS stands for Simple Knowledge Organisation System. It is currently under scrutiny by the W3C
Semantic Web Deployment Working Group and is planned to be published as a W3C Proposed
Recommendation in 2008. See http://www.w3.org/2004/02/skos.
8 In the following “skos:" stands for http://www.w3.org/2004/02/skos/core#.
-
-
Semantic properties are used to represent the structural relationships between
concepts, which are usually at the core of controlled vocabularies like thesauri.
The construct skos:broader denotes the generalization link (BT in standard
thesauri), while skos:narrower denotes its reciprocal link (NT), and
skos:related the associative relationship (RT).
Documentation properties. Often, informal documentation plays an important
role in a KOS. SKOS introduces explanatory notes – skos:scopeNote,
skos:definition, skos:example – and management notes –
skos:changeNote, skos:historyNote etc.
Concept scheme description
A KOS as a whole also has to be represented and described. SKOS coins a
skos:ConceptScheme construct for this. It also introduces specific properties to
represent the links between different KOSs and the concepts they contain. The term
skos:inScheme asserts that a given concept is part of a given concept scheme,
while skos:hasTopConcept states that a KOS contains a concept as the root of
(one of) its constituent hierarchical tree(s), i.e., a concept without a broader concept.
Conversion from a KOS native representation to SKOS RDF data requires the
analysis of the original model of the KOS, and the linking of the elements of this
model to the SKOS ones that fit them most (Assem et al., 2006). One can, for
instance, decide to represent a “class” in a classification scheme as a resource of
type skos:Concept. Based on such a specification, it is then possible to implement an
appropriate conversion program – e.g., an XSL stylesheet when the vocabulary is
natively encoded in XML – to automatically convert the initial representation to a
SKOS one.
As an example, a subject 11F coming from the Iconclass concept scheme 9, “the
Virgin Mary”, identified by the (as yet fictive) resource http://www.iconclass.nl/s_11F,
could be partly represented by the graph in figure 3.
Figure 3 A SKOS graph partly representing the Iconclass subject 11F.
Quoted strings are plain literals. “@” specifies the language of a literal: “en” is the tag
for “English”, “fr” for “French” and “zxx” stands for any “artificial language”.
3. Vocabulary alignment as a solution to the interoperability problem
Having unified and linkable representations of the concepts contained in different
collections’ vocabularies helps managing them in a single framework. However, this
is not sufficient for solving the semantic interoperability problem. One still has to
determine semantic similarity links between the elements of the different
9
See http://www.iconclass.nl.
vocabularies – to align10 them (Doerr, 2001). Fig. 4 illustrates that if a search engine
“knew" that a SKOS concept C from a thesaurus T1 is semantically equivalent to a
SKOS concept D from thesaurus T2, then it could return all the objects that were
indexed against D for a query for objects described using C. The objective is
therefore to align as many concepts of one thesaurus to their semantic equivalents in
the other thesaurus. Where such equivalency cannot be established, it may be
possible to establish links between concepts of one thesaurus and concepts of the
second thesaurus that are either more specific or more general, and to exploit such
“narrower than” and “broader than” relations for query processing.
Figure 4 Using vocabulary alignment for integrated access to different collections
Such an approach has been investigated for subject vocabularies in projects such as
HILT (Macgregor et. al., 2007). The alignment of these vocabularies is however a
labour-intensive task that requires considerable expertise in the concerned thesauri.
Manual alignment has been approached by several projects, notably, CARMEN
(Krause, 2003), Renardus (Day et. al., 2005), KoMoHe11, AOS (Liang & Sini, 2006)
or the ongoing CRISSCROSS12, MACS13 (Landry, 2004) and MSAC (Balikova,
2005). These projects have yielded very interesting results such as the development
of tools to support manual alignment, the deployment of search engines that exploit
resulting alignments, and the contribution of initial methodological ideas. However,
they also demonstrated the complexity, difficulty, and cost of manually aligning large
vocabularies (usually containing many thousand concepts) in realistically-sized
collections and settings. Given that manual labour is expensive and that vocabularies
evolve over time, it is clear that the construction and maintenance of alignment
constitutes an important issue that needs to be addressed. There is a need for
developing advanced, computer-based tools that can identify candidate mappings
between two vocabularies, and that can then propose them to the human expert for
consideration. Alignment would thus become a semi-automatic task where thesaurus
10
Alignment refers in this paper to the creation of semantic relationships (e.g. equivalence) between
concepts coming from different KOSs in order to solve interoperability problems. This notion
approximates what is referred to in the KOS community by vocabulary mapping, crosswalk or
reconciliation, and in the Semantic Web community by ontology alignment, mapping or matching.
11
See http://www.gesis.org/en/research/information_technology/komohe.htm
12
See http://www.d-nb.de/wir/projekte/crisscross.htm
13 See http://macs.cenl.org.
experts’ work would be assisted, and where the integration of collections would
become more cost-efficient.
Recently, the Semantic Web community has produced alignment tools that address
the specific problem of formal ontology matching (Shvaiko & Euzenat, 2005).
However, the techniques they employ and the goals they advertise make them
deployable in a more general context, including thesauri and other similar KOSs.
Although most of the existing ontology alignment tools rely on sophisticated methods
(Euzenat & Shvaiko, 2007), they can be classified and described by the basic
techniques they build upon and the different sources of information they exploit: the
lexical information attached to the concepts of the vocabularies, the structure of
vocabularies, the collection objects described by vocabularies, or other (external)
knowledge sources.
Lexical alignment techniques
In these techniques the lexical materializations of the concepts are compared to each
other. If a significant similarity is found, then we can establish a semantic link
between the concerned concepts. A straightforward example is when two concepts
have the same label. But one can also search for string inclusion patterns or more
complex techniques relying for instance on lemmatizers – getting normalized forms of
labels, e.g. “tree” for “trees” – and syntactical analysis tools. A concept labelled “(map
of) the North Pole” can be detected as a narrower concept of another which is
labelled “Charts, maps”. These lexical methods exploit the preferred labels of
concepts, but they can also turn to their lexical variants or their associated definitions
and scope notes.
Of course, such approaches encounter the same problems as humans when dealing
with words taken out of context. Polysemy and homonymy, for instance, are common
sources of errors. This has to be compensated with contextual information.
Structural alignment techniques
The first kind of context is provided by the vocabulary itself, as it contains hierarchical
and associative links between concepts. These links, especially those concerning
hierarchical generalization and specialization, are useful to constrain a concept’s
natural interpretation: “bank” will be understood differently if it is a narrower term of
“finance” or “geography”. Some tools will analyze this semantic context, either to
check similarities obtained by other techniques or to derive new similarities from
existing ones. If two concepts from different vocabularies are semantically equivalent,
this equivalence will positively influence the alignment tool when it will examine the
children of these concepts to find similarities between them.
Extensional alignment techniques
The second kind of context comes from the actual usage of the concepts in real-life
applications. For instance, a class from a classification scheme will be used to
categorize a number of objects (e.g., books) in a collection. Accessing this
information will provide an “extensional” characterization of the class’ intended
meaning – akin to its literary warrant. When documents are described using two
different vocabularies14, statistical techniques can be employed to compare the sets
of documents described by the concepts from these vocabularies (Figure 5). A high
degree of overlap between these sets will yield a high similarity between
corresponding concepts. Several such techniques have already been experimented
in the KOS field, as in (Zhang, 2006) or (Isaac et. al, 2007).
Figure 5 Using object-level information to align vocabularies
[adapted from (Harmelen, 2005)]
Background knowledge-based alignment techniques
A final group of alignment methods rely on knowledge sources that are external to
the application and the vocabularies being considered. These sources can be of a
different nature as for instance the use of existing general-purpose ontologies like
CYC15 or semantic networks like Wordnet (Miller, 1995). These sources can
contribute KOS-external knowledge to compensate for the lack of KOS-internal
lexical or structural information. For example a concept “calendar” from one
thesaurus can be aligned to the more general concept “publication” from another
thesaurus, using the hypernymy relation that holds between the two corresponding
terms in Wordnet (Fig.6).
14
This also applies to the more general case when the similarity between objects from two collections
described by their own vocabularies can be assessed, applying for example text similarity measures
on textual documents.
15 See http://www.opencyc.org.
Figure 6 Using background knowledge to align vocabularies
[adapted from (Harmelen, 2005)]
4. Integrated Collections Access: an Example
To illustrate the potential of the described technology, we used it for creating an
integrated access to two collections in two different Dutch cultural heritage
institutions, the Rijksmuseum, and the National Library of the Netherlands (Gendt et
al., 2006). The Manuscripts collection contains 10,000 medieval illuminations which
are annotated by subject indices describing the content of the image. These indices
come from the Iconclass classification scheme, a vocabulary of 25,000 elements
designed for iconographical analysis. The Masterpieces collection contains 700
objects such as paintings and sculptures and its subjects are indexed using the ARIA
“catalogue”, a vocabulary conceived mainly as a resource for hierarchical browsing.
Both vocabularies were translated into SKOS, and mappings between them were
calculated with existing state-of-the art mapping tools, namely, Falcon (Jian et al.,
2005) and S-Match (Giunchiglia et al., 2005). Falcon uses a mixture of lexical and
structural techniques. In addition to lexical techniques, S-Match uses Wordnet as
background knowledge, and exploits “semantic reasoning” using a logical
interpretation of the concepts based on the structure of the vocabularies.
We implemented a faceted browser, in which the mappings and the vocabularies’
semantic web representations are exploited to provide integrated assess to the
collections, offering three different views: single, combined, and merged view.
The Single View presents the integrated collections from the perspective of just one
of the vocabularies. In the screen capture (Fig. 7) the first four pictures come from the
Rijksmuseum, the others are Illuminated Manuscripts. Browsing is done solely using
the ARIA Catalogue, i.e. these illuminations have been selected exploiting the
mapping between the currently selected ARIA concept “Animal Pieces” and the
Iconclass concept “25F:animals”.
Figure 7 Single View: Using the ARIA thesaurus to browse the two collections [from (Gendt
et al., 2006)]
The Combined View provides simultaneous access to the collections through their
respective vocabularies in parallel. This allows us to browse through the integrated
collections as if it was a single collection indexed against two vocabularies. In figure
8, we made a subject refinement to ARIA “Animal pieces”, and narrowed down our
search with Iconclass to the subject “Classical Mythology and Ancient History”.
Figure 8 Combined View: Using ARIA and Iconclass to browse the two collections [from
(Gendt et al., 2006)]
Finally, the Merged View gives access to the collections through a merged thesaurus
combining both original vocabularies into a single facet, based on the links found
between them in the automatic mapping process. If we select the ARIA concept
“Animal pieces”, the view provides both ARIA concepts such as “Birds” and Iconclass
concepts such as “29A:animals acting as human beings” for further refining our
search.
5. Discussion and conclusion
Existing alignment tools have been reported to perform relatively poorly on real-life
cases such as cultural heritage thesaurus alignment (Gendt et al., 2006). In fact,
alignment is still an open research problem as no single technique is universally
applicable, or will return satisfactory results. In practice, different techniques have to
be carefully selected and combined, depending on the characteristics of the case at
hand, such as the richness of the semantic structures of vocabularies, their lexical
coverage and the existence of collections simultaneously described by several
vocabularies. It should be noted, however, that a continuous improvement of
techniques and tools can lead to significant improvements, as witnessed in the
regular evaluation campaigns organized by the research community (Euzenat et al.,
2006).
The Semantic Web-inspired methods and tools described in this paper still require
further experimentation in practical applications, and a greater availability of
vocabularies. Nevertheless, the existence of current representation and alignment
techniques already allows the creation demonstrators showing their potential value
for integrating collections at the semantic level, leading from separate islands of
cultural heritage knowledge to better connected networks of collections and
vocabularies.
One such demonstrator is described in Section 4; this faceted browser gives a unified
access to two collections of illuminated manuscripts via any of its respective
metadata descriptions. Other examples of Web portals that illustrate the use of
Semantic Web techniques in the cultural heritage domain can be seen on the
websites of the MuseumFinland16 and eCulture17 projects. These projects, even if not
focusing on semantic alignment, demonstrate the possible benefits of using Semantic
Web technologies: the use of the SKOS representation format, the development of
innovative interfaces to access Cultural Heritage collections, and the exploitation of
automated reasoning techniques over RDF-based metadata.
Other portals are being created with enhanced functionality and usability, as the
synergy between CH and SW communities increases: one example is the ongoing
eCulture project, which has been given the Semantic Web Challenge18 award in
2006. In fact, the richness and high quality of cultural heritage data is very attractive
to researchers of the Semantic Web community, as they have many tools but little
real-life metadata to show their true potential. On the other hand, the CH domain
(including digital libraries) could profit from techniques and tools developed by the
SW community in creating a Web of cultural heritage that delivers high quality
content via easily accessible semantic search and navigation. Semantic search –
matching meanings – represents a huge advance in relation to current web searching
techniques that are based on full text search – matching strings.
References
Antoniou, G., Harmelen, F. van (2004) Semantic Web Primer. Cambridge, MA: MIT
Press.
Assem, M. van, Malaise, V., Miles, A., Schreiber, G. (2005) A Method to Convert
Thesauri to SKOS. 3rd European SemanticWeb Conference, Budva, Montenegro,
2005.
Balikova, M. (2005) Multilingual Subject Access to Catalogues of National Libraries
(MSAC) Czech Republic’s collaboration with Slovakia, Slovenia, Croatia, Macedonia,
Lithuania and Latvia. Paper presented at the 71th IFLA General Conference and
Council "Libraries - A voyage of discovery", August 14th - 18th 2005, Oslo, Norway.
Available at: http://www.ifla.org/IV/ifla71/papers/044e-Balikova.pdf.
Berners-Lee, T., Hendler, J., Lassila, O. (2001) The Semantic Web. Scientific
American, 284 (5), 34-43.
Available at: http://www.sciam.com/2001/0501issue/0501berners-lee.html.
Brickley, D., Guha, R. V., Editors (2004) RDF Vocabulary Description Language 1.0:
RDF Schema. W3C Recommendation, 10 February 2004. Latest version available at
http://www.w3.org/TR/rdf-schema/
Day, M., Koch, T., Neuroth, H. (2005) Searching and browsing multiple subject
gateways in the Renardus service. In Proceedings of the Sixth International
Conference on Social Science Methodology, Amsterdam, 2005.
Doerr, M. (2001) Semantic Problems of Thesaurus Mapping. Journal of Digital
Information, 1 (8).Article No. 52.
See http://www.museosuomi.fi.
See http://e-culture.multimedian.nl.
18
See http://challenge.semanticweb.org.
16
17
Available at: http://jodi.tamu.edu/Articles/v01/i08/Doerr/.
Euzenat, J. et al. (2006) Results of the Ontology Alignment Evaluation Initiative 2006.
International Workshop on Ontology Matching, 5th International Semantic Web
Conference (ISWC 2006), Athens, Georgia, USA.
Available at:http://www.dit.unitn.it/~p2p/OM-2006/7-oaei2006.pdf.
Euzenat, J., Shvaiko, P. (2007) Ontology Matching. Berlin, Heidelberg: Springer.
Gendt, M. van et al. (2006) Semantic Web Techniques for Multiple Views on
Heterogeneous Collections: a Case Study. In: Julio Gonzalo et al. (Eds.). Research
and advanced technology for digital libraries: proceedings of the10th European
Conference on Research and Advanced Technology for Digital Libraries (ECDL
2006), Alicante, Spain, September 17-22 2006. Berlin, Heidelberg: Springer. (Lecture
Notes in Computer Science, 4172), 426-437.
Giunchiglia, F., Shvaiko, P., Yatskevich , M. (2005) Semantic Schema Matching. 13th.
International Conference on Cooperative Information Systems (CoopIS 2005).
Harmelen, F. van (2005) Ontology Mapping: A Way Out of the Medical Tower of
Babel? In: S. Miksch, J. Hunter and E. Keravnou (Eds.). Artificial Intelligence in
Medicine: proceedings of the 10th Conference on Artificial Intelligence in Medicine,
AIME 2005, Aberdeen, UK. Berlin, Heidelberg: Springer. (Lecture notes in computer
science, 3581), 3-6.
Isaac, A., Meij, L. van der, Schlobach, S., Wang, S. (2007) An empirical study of
instance-based ontology matching. Proceeedings of the 6th International Semantic
Web Conference (ISWC 2007). Busan, Corea.
Jian, N., Hu, W., Cheng, G., and Qu, Y. (2005) Falcon-AO: Aligning Ontologies with
Falcon. K-CAP Workshop on Integrating Ontologies, Banff, Canada, 2005.
Krause, J. (2003) Standardization, heterogeneity and the quality of content indexing:
a key conflict of digital libraries and its solution. World library and information
congress: 69th IFLA general conference and council.
Landry, P. (2004) Multilingual Subject Access: The Linking Approach of MACS.
Cataloging & Classification Quarterly, 37 (3-4), 177-191.
Liang, A., Sini, M. (2006) Mapping AGROVOC and the Chinese Agricultural
Thesaurus: Definitions, tools, procedures. New Review in Hypermedia and
Multimedia, 12 (1), 51- 62.
McGuinness, D. L., Harmelen F. van, Editors (2004) OWL Web Ontology Language
Overview. W3C Recommendation, 10 February 2004. Latest version available at
http://www.w3.org/TR/owl-features/
Miles, A., Brickley, D. (2005) SKOS Core Guide. W3C Working Draft. Work in
progress, latest version available at http://www.w3.org/TR/swbp-skos-core-guide/
Miller, G. (1995) Wordnet: a lexical database for English. Communications of the
ACM, 38 (11), 39-41.
Macgregor, G., McCulloch, E., Nicholson, D. (2007) A DDC spine based terminology
server for improved resource discovery, 2nd International Conference on Metadata
and Semantics Research. Corfu, Greece, 2007.
Shvaiko, P., Euzenat, J. (2005) Ontology Matching. D-Lib Magazine, 11 (12), In Brief.
Available at: http://www.dlib.org/dlib/december05/12inbrief.html
Zhang, X. (2006) Concept integration of document databases using different indexing
languages. Information Processing and Management, 42, 121–135