=Paper=
{{Paper
|id=None
|storemode=property
|title=Crowdsourced Semantics with Semantic Tagging: "Don't just tag it, LexiTag it!"
|pdfUrl=https://ceur-ws.org/Vol-1030/paper-09.pdf
|volume=Vol-1030
|dblpUrl=https://dblp.org/rec/conf/semweb/Veres13
}}
==Crowdsourced Semantics with Semantic Tagging: "Don't just tag it, LexiTag it!"==
https://ceur-ws.org/Vol-1030/paper-09.pdf
Crowdsourced Semantics with Semantic Tagging: “Don’t
just tag it, LexiTag it!”
Csaba Veres
Institute for Information and Media Science,
University in Bergen, Norway
Csaba.Veres@infomedia.uib.no
Abstract. Free form tagging was one of the most useful contributions of
“Web2.0” toward the problem of content management and discovery on the
web. Semantic tagging is a more recent but much less successful innovation
borne of frustration at the limitations of free form tagging. In this paper we
present LexiTags, a new platform designed to help realize the potential of
semantic tagging for content management, and as a tool for crowdsourcing
semantic metadata. We describe the operation of the LexiTags semantic
bookmarking service, and present results from tools that exploit the semantic
tags. These tools show that crowdsourcing can be used to model the
taxonomy of an information space, and to semantically annotate resources
within the space.
Keywords. crowdsourcing, metadata, bookmarking, tagging, semantic tags
1 Introduction
The emergence of "Web2.0"1 brought a number of innovations which changed the
way people interact with information on the World Wide Web. The new paradigms made it
easy for anyone to contribute content rather just consume. One of the early success stories
was social tagging, which gave rise to folksonomies 2 as a way to organise and find
information on the Web through emergent shared vocabularies developed by the users
themselves. Social tagging for content management and discovery became very popular in
commercial services like the photo sharing site flickr.com and the bookmarking site
delicious.com. These successes prompted some commentators to declare victory of user
driven content tagging over the “overly complex” technologies of the semantic web.
Perhaps most famously, in a web post entitled “Ontology is Overrated: Categories, Links,
and Tags” Clay Shirky argued that any technology based on hierarchical classification
(including ontologies) was doomed to fail when applied to the world of electronic resources
[1]. Instead, simple naive tagging opened the door to crowdsourced content management,
where dynamic user contributed metadata in a flat tag space offered a breakthrough in
findability.
However, researchers and information architects soon began to point out the
limitations of unconstrained tagging for enhancing information findability. [2] identified a
number of problems with tagging, which can limit its effective usefulness. Among the
problems were tag ambiguity (e.g. apple - fruit vs. apple - company), idiosyncratic
1 T. O'Reilly. What Is Web 2.0: Design Patterns and Business Models for
the Next Generation of Software. http://www.oreillynet.com/pub/a/oreilly/tim/news/
2005/09/30/what-is-web-20.html, 2005.
2 http://iainstitute.org/news/000464.php#000464
treatment of multi word tags (e.g. vertigovideostillsbbc, design/css), synonyms (e.g. Mac,
Macintosh, Apple), the use of acronyms and other terms as synonymous terms (e.g. NY,
NYC, Big Apple), and of course mis spelled and idiosyncratic made up tags. These factors
limit the use of tags in large scale information management. For example [3] discuss
limitations of searching with tags, which is necessarily based on syntactic matching since
there are no semantic links between individual tags. Thus, searching with “NYC” will not
guarantee that results tagged with “NY” will be retrieved.
Semantic tags or rich tags as they are known in the context of social tagging,
emerged as a way to impose consistent and refined meanings to user tags [4]. The two best
known semantic tagging sites were Faviki 3 and Zigtag4 (the latter now appears to be
defunct). Each site expected users to use tags from a large collection of provided terms.
Faviki used WikiPedia identifiers, while Zigtag used a “semantic dictionary”. Both sites
also allowed tags which were not in their initial knowledge base. In the case of Faviki,
users can link undefined tags to a web page which best represents the tag. The web page is
located by a simple web search. Zigtag allowed the use of undefined tags, with the
expectation that users would later return and provide definitions of the tags. However a
large proportion of tags remained without definition, leading to a mishmash of defined and
undefined tags.
LexiTags [5] was initially developed for similar reasons, as a tool for content
management with rich tags. But as a semantic application it also had higher aspirations, to
provide a platform and a set of front end tools designed to crowdsource the semantic web.
By providing intuitive access points (APIs) to user generated metadata with semantic tags,
the platform was expected to outgrow its initial purpose and provide novel new benefits for
its users while at the same time generating semantic metadata for general consumption. The
intuition is that users should have systems which behave as Web2.0 at the point of insertion,
yet as Semantic Web at the point of retrieval. In this paper we present the core LexiTags
system, and describe some tools we have developed to capitalise on the crowdsourced
semantic metadata.
2. LexiTags
LexiTags5 is an acronym for “Lexical Tags”, from the fact that the tags are
primarily lexical items, or natural language dictionary words. They are disambiguated
through the use of an interface which presents the user with a set of choices from WordNet,
an electronic lexical database [6]. The main content bearing units in WordNet are synsets,
which are represented by contextually synonymous word meanings grouped in a single
entry. The word couch for example is represented by the synset {sofa, couch, lounge}. But
couch is of course an ambiguous word whose alternate meaning appears is a second synset
{frame, redact, cast, put, couch} as in “Let me couch that statement …. “. There is
therefore no ambiguity in WordNet because every synset represents a unique meaning for a
word string. LexiTags presents such synsets, and a short gloss, to help users chose their
intended meaning.
The use of synsets as tags can combine precise definitions without the need to
adopt a set of idiosyncratic keywords. Mappings can be set up between synsets and any
other vocabulary, enabling specific keyword markup through a natural language interface.
3 http://www.faviki.com/pages/welcome/
4 http://zigtag.com
5 http://lexitags.dyndns.org/ui/webgui/
The success of mapping efforts can of course vary. While highly technical ontologies could
prove difficult, lightweight ontologies and taxonomies like schema.org are not
problematical, and in section 3.2 we will see a tool that makes use of exactly such a
mapping.
Fig. 1 shows a detail of the main interface of LexiTags, which is mainly a simple
list of URLs that have been bookmarked. Clicking on the URL opens a new window with
the web site. The user tags appear below the URL. Hovering the mouse over these tags
pops up a definition which clarifies the precise sense of that tag. Clicking on a tag will open
a new tab which shows bookmarks tagged with the same tag sense.
Fig. 1. The main LexiTags interface
Tags for a bookmark are entered freely in the text box near the bottom of the “Edit
Bookmark” window (fig. 2) which is popped up through the use of a bookmarklet. The user
types one tag at a time into the text box and presses enter which puts each tag in a list
above the text box, initially in red colour. Users can simply enter tags as they like, as in
most other tagging sites. Finally, users click on each undefined tag to add disambiguation
through the final “Editing the tag …” window, shown in figure 3.
Fig. 2. Window for adding and editing tags
The “Edit tag ..” window shows the possible interpretations from WordNet, or
DBPedia if WordNet does not have an entry for the tag word. This is the case most often
when the tag is the name of a company or a person or a new technology. DBPedia also
includes a large number of common abbreviations, such as “NYC”. In addition, DBPedia
defines mappings to WordNet synsets for many concepts, which helps fill gaps in WordNet.
Unfortunately the coverage is not complete, so “NYC” for example is not linked to the
synset for New York in WordNet. Users must select the sense that best matches their intent.
The choices are ordered by word frequency, and our experience suggests that the intended
sense is amongst the first two or three senses. Since the main point of tagging is for future
retrieval, it would make sense if people tended to avoid words in their obscure, low
frequency senses. However this is just conjecture at the moment, and we are investigating
other methods for optimal ordering. One approach is to weight the rankings according to
the aggregate distance of each candidate sense to a context tag provided by already
disambiguated tags. Another approach is to personalise the rankings so that each user’s own
tagging history influences the ranking of the candidate senses.
As each tag is disambiguated by the user, it turns green. Any tag left
disambiguated is deleted when the user presses the “OK” button, so every tag in the
LexiTags platform is a disambiguated, defined term.
Fig. 3. Disambiguation window
3. The Lexitags ecosystem
If LexiTags were just a bookmarking service with rich tags, there would be little to
differentiate it from Zigtags. But the idea was to use the bookmarked sites and their tags as
a starting point for a set of tools that extracted value from the tags. As such, LexiTags
should be seen as a platform to expose crowdsourced semantic metadata to clients, both for
creation and consumption of metadata. In terms of content creation clients, we are
developing an iPhone app for tagging photographs with LexiTags, as described in [5]. Due
to space limitations we are unable to discuss alternative input applications, but instead
describe two applications for consuming the metadata. One creates a content taxonomy, the
other produces metadata for the web.
3.1 Content taxonomy
[7] discuss SynsetTagger6 , which was developed to consume LexiTags tags
automatically, but has thus far only been demonstrated in manual mode to create
lightweight ontologies from user input. SynsetTagger makes use of select WordNet
6 http://csabaveres.net/csabaveres.net/Semantic_Apps.html
relations to construct a lightweight ontology by inferring additional nodes from the
provided tags. The most important link for nouns is hyponymy/hypernymy which are the
semantic relations otherwise known as subordination/superordination, subset/superset, or
the IS-A relation. A concept represented by the synset {x, xʹ′, . . .} is said to be a hyponym of
the concept represented by the synset {y, yʹ′, . . .} if native speakers of English accept
sentences constructed from such frames as ”an x is a (kind of) y” [8], [9]. Another relation
used by SynsetTagger is meronymy, or the part/whole relation. A concept {x,x’….} is a
meronym of a concept {y, yʹ′, . . .} if “an x is a part of y”, and it is a holonym if “a y is a part
of x”. There are several other important relations in WordNet, some of which will be
mentioned in the case study.
SynsetTagger works by constructing the hypernym chain and pruning nodes if
their information content is trivial, which is determined by counting the outward edges
from each node and eliminating each node that falls below a threshold value. For example,
fig. 4 shows a taxonomy constructed from the input synsets coloured green. The orange
coloured inferred hypernyms will be discarded because the have only a single outgoing
edge (or are too close to the top level node), and the clear white ones will be kept. The grey
nodes are both inferred and asserted, and will also be kept in the final taxonomy. The nodes
that are kept are called informative subsuming nodes.
Fig. 4. Complete hypernym chain constructed from input synsets
The tool has a number of user configurable parameters which determine the final
selection of nodes. Two important ones are the number of outgoing edges, and the distance
from the topmost node. The optimal selection is a matter of trial and error with a given
input set. The pruning mechanisms are similar to the “nearly automated” approach to
deriving category hierarchies for printed text [10], but SynsetTagger differs in that it allows
users to adjust the parameters and receive immediate visual feedback about their
consequences on the constructed taxonomy. It is intended as an interactive tool to give
users a sense of control over their content.
3.2 Metadata for the Web
MaDaME (Meta Data Made Easy) is a tool for embedding semantic metadata into
web sites. Its development was spurred by the release of the schema.org initiative, which is
a type schema meant to be used by web masters to add structured metadata to their content.
The incentive for web masters to use the schema is that web sites that contain markup will
appear with additional details in search results, which enable people to judge the relevance
of the site more accurately and hopefully increase the probability that the site will be
visited.
Clearly this is an important development in the effort to crowdsource semantic
web content. However, the schema was designed specifically for the use case of search, and
both the semantics and the preferred syntax reflect that choice. In terms of semantics, the
schema contains some non-traditional concepts to fulfil its intended use. For example there
is a general class of Product but no general class for Artifact. There are also odd property
ascriptions from the taxonomy structure, so, for example, Beach has openingHours and
faxNumber. In terms of syntax, there is a very strong message that developers should use
the relatively new microdata format rather than the more popular RDFa web standard[11].
This is unfortunate because it makes metadata from schema.org incompatible with many
other sources of metadata like Facebook’s OGP 7.
A strong motivation for MaDaME was to create a tool that would not only help
web designers apply schema.org markup to their web sites, but to simultaneously inject
ontology terms from other standard sources into their web sites in the RDFa standard. In
other words, to maximise the crowdsourcing potential offered by schema.org. This was
achieved through mappings between WordNet and schema.org, as well as SUMO [12]. In
normal operation users select key words in their web sites, which are then disambiguated
using the LexiTags interface. The WordNet synsets are stored, and their most appropriate
mapping to schema.org and SUMO computed. These are then inserted into the html source,
and made available to the web designer for further refinement.
While MaDaME is currently presented as a standalone tool8 it can also be used to
automatically annotate any web site bookmarked on LexiTags. These annotations can be
sent to the maintainers of the web sites with a cover letter explaining the purpose of the
markup. Alternatively, the markup could be stored on the LexiTags platform and offered
through an API.
4. Results
We present a short evaluation of SynsetTagger on a set of approximately 100
bookmarks for a single user on LexiTags, and then the metadata generated for one web site
on MaDaME.
Fig. 5 shows a portion of the taxonomy generated from the semantic tags used on
the set of 100 bookmarks. The asserted green tags are used to build the hypernym chain
from WordNet. If any node is selected in the interface, the set of connected nodes will also
be highlighted, making it easier for users to understand the relationships in the taxonomy.
For example the asserted tag conference can be seen as a kind of meeting which in turn is a
kind of gathering, and so on. The orange nodes are inferred hypernyms, but they will be
rejected in the final taxonomy for one of two reasons; a) they have fewer than the specified
number of children, or b) they are closer than the selected cut from the entity node.The
white nodes are the inferred nodes which will be retained in the final export. Users can
therefore manipulate the two parameters until the desired level of generality is reached.
We will see that in general a lower number of requisite children and a lower cut
will result in fewer nodes. This may be counter intuitive at first, but the rather
straightforward explanation is that more and more asserted tags fail to find a suitable
subsuming concept when the criterion for retaining the concepts becomes more stringent.
7 http://ogp.me
8 http://csaba.dyndns.ws:3000
Fig. 5. Taxonomy from approx. 100 bookmarked sites
Fig. 6 shows the hyponyms of entity that are retained with “children” set at two
and “cut” also set at two. The black arrow mark at the right side of each oval shows that the
node can be extended to reveal more children. Every tag except for weather found a more
general subsuming hypernym when the criterion for subsuming nodes is lax. The tags
which do not have a subsuming concept are noteworthy because they represent unusual
bookmarks which stand apart from the rest. The subsuming concepts themselves are very
general in this example, and their utility for browsing the bookmarked resources is
questionable.
Fig. 6. Inferred hypernyms with lax criteria
Fig. 7 shows the same set of tags with “children” set at 5. Since the criterion for
subsuming concepts is much higher, there are many more tags which do not have a
subsumer. For example the two tags fund and date are both a kind of measure in fig. 6. But
these are the only two kinds of measure in the tag set, so with a criterion of 5 children,
measure is no longer considered as an informative subsuming node. On the other hand the
remaining subsuming concepts have differentiated and are now somewhat more specific.
For example group is replaced with the more specific organization which is one of only
three kinds of group in the tree. So group is discarded but organization is kept because
there are many different kinds of organization in the tree. Overall the constructed taxonomy
is much more useful for browsing because the general categories are now more informative,
and the outliers are not forced into overly general categories.
Fig. 7 Inferred hypernyms with stringent criteria
Fig. 8. shows that expanding the artifact node reveals some useful sub categories
to browse. In general, users can easily configure the parameters to arrive at an optimal
taxonomy for their needs.
Fig. 8. Different kinds of artefacts in the bookmark set
It is at this point where LexiTags becomes a system that behaves as Web2.0 at the
point of insertion, yet as Semantic Web at the point of retrieval. Since the tags have precise
definitions they avoid the pitfalls of free form tags like vagueness and ambiguity, and
because the defined tags participate in various relations they add value to the asserted tags.
They become self organising, with a neat hierarchical structure emerging automatically.
Even in this small example of 100 bookmarks and around 400 tags the emerging
generalisations form a useful browsing hierarchy. We expect the categorisations to improve
with more bookmarks and tags, resulting in a situation where a growing set of bookmarks
leads to more organisation rather than complete chaos as seen in traditional free form
tagging systems. The users must gain such benefits because they are asked to perform a
little more work at the point of insertion. An important point of our work is to enhance
LexiTagging services to provide additional benefit to the users, to encourage them to
contribute semantic metadata.
The emergent taxonomy can be extended with additional related terms from
WordNet. SynsetTagger already has the functionality to add some additional relations to
enrich the ontology. Currently these include the two part-of relations, and the domain
terms. For example movie has meronyms episode, credits, subtitle, caption, scene, shot, and
domain terms dub, synchronize, film, shoot, take, videotape, tape, reshoot. These can all be
added with the appropriate relations. In addition the synonyms appearing in each synset
could also be included. Sometimes this is quite rich, as is the case in our example of movie,
whose synset consists of {movie, film, picture, moving picture, moving-picture show,
motion picture, motion-picture show, picture show, pic, flick}. WordNet also provides a set
of coordinate terms which are nouns or verbs that have the same hypernym as the target
word. Once again movie has a number of coordinate terms including stage dancing,
attraction, performance, burlesque, play, and variety show.
A final source of data is user tags which are adjectives or verbs, and were not used
in the construction of the taxonomy. These tend to be descriptive words that are suitable for
use as properties, e.g. Hungarian, fine_tune, synchronize, semantic, amusing, and open.
When all of this extra information is added to the taxonomy it results in a greatly enriched
ontology which can be used to provide additional services like matching users against one
another, or to aid content discovery and recommendation.
As part of the crowdsourcing effort we are planning to enable the export of the
generated ontologies and the URLs to which they apply. This would include a list of topics
and descriptions where available, as well as relations to other topics of interest. The
semantic metadata for each URL would be stored on our servers and made available
through an API.
The second form of metadata creation makes use of other vocabularies that have
been mapped to WordNet as in the mapping tool MaDaME, which can contribute
schema.org and SUMO metadata for any bookmarked URL.
Consider for example one URL bookmarked in the LexiTags data, http://
www.imdb.com. This site was tagged with trailers, information, and movies. When
submitted to MaDaME, it can automatically generate markup that can be inserted into the
HTML site, or provided as additional data through the aforementioned information API.
The mappings generated from the tags are shown in fig. 9. Note that the is
currently assigned to random words at the beginning of the text, and this needs to be
inserted into an appropriate location in the original HTML if it is used to mark up the page
directly. The example shows that MaDaME currently generates markup from three
vocabularies. The original WordNet synset is preserved, as well as the corresponding
SUMO class. The SUMO class mappings tend to be quite precise because there is an
extensive set of mappings readily available for SUMO9. On the other hand the mappings to
schema.org are significantly more sparse, because the schema.org types are considerably
fewer in number. In cases where no exact match is found a heuristic procedure is used to
determine the closest match, and this can result in overly general or erroneous mappings.
For example trailer, preview is mapped to schema:Intabgible whereas a more appropriate
mapping might be to the concept schema:VideoObject (CreativeWork > MediaObject >
VideoObject). On the other hand, many concepts simply do not have a more precise
mapping in schema.org, and the mapping of information, data to schema:Intangible
appears to be correct.
9 http://sigma-01.cim3.net:8080/sigma/Browse.jsp?
kb=SUMO&lang=EnglishLanguage
NewsDesk
movie
familiar
Fig. 9. Automatic schema.org and SUMO annotation for imdb.com
The key point is that metadata from various different namespaces can
automatically be made available for different consumers, simply as a side effect of
bookmarking. Search engines can use the schema.org markup, while other services can use
the SUMO or WordNet classifications. If the bookmarking service were to grow in
popularity, it could become a large repository of schema.org markup for a large set of
URLs. Search engines could, presumably, use this markup as if it was embedded in the web
sites themselves. But MaDaME also provides a user interface that can be used to greatly
enhance the schema.org markup using a drop down form as shown in fig. 10. The form
includes a text box for all the possible properties of the schema type which is selected for a
word in the text. Users could select words in addition to the tags already assigned, and add
these to the schema. The figure also shows the extended markup generated by the tool for
the word movie.
movie
Fig 10. Drop down forms and the extensive schema.org markup they can generate
5. Related Work
WordNet is an extremely highly cited resource in all language related areas of
study. The official web site at Princeton University maintains a list of publications10 based
10 http://wordnet.princeton.edu/wordnet/publications/
on WordNet, but this is no longer maintained because it was “growing faster than it was
possible to maintain”.
Within the Semantic Web community WordNet has enjoyed a duality with some
researchers criticising its use as an ontology [13]-[15] while others embracing it either as a
core taxonomy [16] or as a way to infer semantic relations (e.g. [17], [18]).
[10] used WordNet to automatically infer hierarchical classifications in textually
annotated images, and [19] uses it to implement hierarchical faceted classification of
recipes and medical journal titles. Both systems use automated extraction and
disambiguation of key input terms, which differs from our approach where we ask users to
supply these terms. But they use a very similar pruning algorithm to establish the final
taxonomic structure.
The idea that free form user tags can be semantically enhanced has received a
great deal of attention. Most of the existing work focuses on automatically enriching the
tags already present, by exploiting the statistical regularities in the way tags are assigned to
resources by users. [20] suggests that the efforts can broadly be classified as (a) extracting
semantics of folksonomies by measuring relatedness, clustering, and inferring subsumption
relations or (b) semantically enriching folksonomies by linking tags with professional
vocabularies and ontologies, for example Wikipedia, and WordNet [21]-[23]. These
resources are used in various ways, including to effectively cluster tags, for disambiguation,
adding synonyms, and linking to annotated resources and ontology concepts. During this
process the terms of the folksonomy are cleaned up and disambiguated, linked to formal
definitions and given properties which make them more useful as ontologies.
There are also a few studies in which users are expected to contribute semantics at
the time of tagging. [24] studies a corporate blogging platform which included a tagging
interface. The tagging interface was linked to a domain ontology, and whenever someone
typed a tag that had interpretations in the ontology the interface would present a choice of
possible concepts to link the tag to. The ontology would also evolve as users typed new
tags which were initially not in the ontology, but the scope of defined tags was limited by
the ontology. [25] discuss a sophisticated Firefox plugin, Semdrops, which allowed users to
annotate web resources with a complex set of tags including category, property, and
attribute tags. These were aggregated in a semantic wiki of the user’s choosing. [26] reports
on an open source bookmarking application (SemanticScuttle) that has been enhanced with
structurable tags, which are tags that users can enhance with inclusion and equivalence
relations at the time of tagging. [27] describes extreme tagging in which users can tag other
tags, to provide disambiguation and other relational information about tags.
These latter approaches require users to learn new ways of tagging, which are
often more complex and opaque than free form tags. The benefit of LexiTagging is that the
process is minimally different from activities they are already comfortable with. They
simply sign up to a bookmarking site, install a bookmarklet and start tagging. The only
addition to the workflow is to disambiguate tags, but this process is so similar to looking up
definitions in a dictionary that it needs no explanation.
6 Conclusion
The LexiTags platform is a familiar bookmarking platform, like delicious.com,
where users can store the URLs of interesting web sites and tag them with meaningful
terms that aid in successive recall and discovery. The only modification is that the user tags
are simple dictionary words, not disambiguated strings. But this small change gives the
resources on the platform a sound semantic grounding which can significantly enhance the
functionality of the service. Some examples of benefits to users are automatic content
classification and browsing, external content recommendation, enhanced content discovery,
and user profile matching. Some of these services represent crowdsourcing solutions to
existing problems which are difficult to fully automate. For example we have shown how
the lexitags can be used to infer schema.org and SUMO classifications for each
bookmarked web site, which is a task that would otherwise be done manually.
The vision is to create an integrated platform where users begin by simply
bookmarking web sites, but then automatically receive the benefits of the enhanced services
already described. This gives them the incentive to invest in the added effort to
disambiguate their tags. Many of the components are in place, but they need some
programming effort to complete the integration. This paper presented the theoretical
motivation behind the work, and some preliminary results to show what is possible.
7 References
[1] C. Shirky, “Ontology is Overrated--Categories, Links, and Tags,” http://
www.shirky.com/writings/ontology_overrated.html, 2007.
[2] A. Mathes, “Folksonomies-cooperative classification and communication through
shared metadata,” Computer Mediated Communication, 2004.
[3] A. Sheth and K. Thirunarayan, Semantics-empowered Data, Services, and Sensor and
Social Webs. Morgan & Claypool Publishers, 2012.
[4] H. Hedden, “How SEMANTIC TAGGING Increases Findability,” EContent magazine,
08-Oct-2008.
[5] C. Veres, “LexiTags: An Interlingua for the Social Semantic Web,” presented at the
Alexandre Passant,Sergio Fernández,John Breslin,Uldis Bojārs, (Eds.)
Proceedings of the 4th International Workshop on Social Data on the WebIn
conjunction with the International Semantic Web Conference (ISWC2011), Bonn,
2011.
[6] C. Fellbaum, WordNet: An electronic lexical database. Cambridge, MA.: MIT Press,
1998.
[7] C. Veres, K. Johansen, and A. Opdahl, “SynsetTagger: A Tool for Generating Ontologies
from Semantic Tags,” presented at the Proceedings of the 3rd International
Conference on Web Intelligence, Mining and Semantics., 2013.
[8] G. A. Miller, “WordNet: a lexical database for English,” Communications of the ACM,
vol. 38, no. 11, Nov. 1995.
[9] G. A. Miller, R. Beckwith, C. Fellbaum, D. Gross, and G. A. Miller, “Introduction to
wordnet: an on-line lexical database,” CSL Report. International journal of
lexicography, 1993.
[10] E. Stoica and M. A. Hearst, Nearly-automated metadata hierarchy creation.
Association for Computational Linguistics, 2004, pp. 117–120.
[11] P. Mika and T. Potter, “Metadata Statistics for a Large Web Corpus,” LDOW2012,
April 16, 2012, Lyon, France, 16-Apr-2012. [Online]. Available: http://
events.linkeddata.org/ldow2012/papers/ldow2012-inv-paper-1.pdf. [Accessed: 11-
Jul-2012].
[12] I. Niles and A. Pease, “Towards a Standard Upper Ontology,” presented at the
Proceedings of the international conference on Formal Ontology in Information
Systems - FOIS '01, New York, New York, USA, 2001, vol. 2001, pp. 2–9.
[13] A. Gangemi, N. Guarino, C. Masolo, and A. Oltramari, “Restructuring wordnet's top-
level,” AI Magazine, 2002.
[14] A. Gangemi, N. Guarino, C. Masolo, and A. Oltramari, “Sweetening WORDNET with
DOLCE,” AI Magazine, vol. 24, no. 3, p. 13, Sep. 2003.
[15] A. Oltramari, A. Gangemi, and E. al, “Restructuring WordNet's top-level: The
OntoClean approach,” LREC2002, 2002.
[16] F. Suchanek, G. Kasneci, and G. Weikum, “Yago: A Large Ontology from Wikipedia
and Wordnet,” Web Semantics: Science, Services and Agents on the World Wide
Web, vol. 6, pp. 203–217, 2008.
[17] T. H. Duong, T. N. Ngoc, and G. S. Jo, “A Method for Integration of WordNet-Based
Ontologies Using Distance Measures,” KNOWLEDGE-BASED INTELLIGENT
INFORMATION AND ENGINEERING SYSTEMS Lecture Notes in Computer
Science, 2008, Volume 5177/2008, Mar. 2008.
[18] J. Kietz and A. Maedche, “A method for semi-automatic ontology acquisition from a
corporate intranet,” Workshop “Ontologies and text, 2000.
[19] E. Stoica and M. Hearst, “Demonstration: Using wordnet to build hierarchical facet
categories,” presented at the ACM SIGIR Workshop on Faceted Search (August
2006)
[20] F. Limpens, F. Gandon, and M. Buffa, “Linking Folksonomies and Ontologies for
Supporting Knowledge Sharing,” Projet ISICIL :Intégration Sémantique de
l“Information par des Communautés d”Intelligence en LigneAppel ANR
CONTINT 2008 ANR-08-CORD-011-05, 01-Aug-2009. [Online]. Available: http://
isicil.inria.fr/v2/res/docs/livrables/ISICIL-ANR-EA01-
FolksonomiesOntologies-0906.pdf. [Accessed: 15-Aug-2011].
[21] L. Specia, “Integrating folksonomies with the semantic web,” The semantic web:
research and applications, 2007.
[22] Angeletou, Sofia; Sabou, Marta; Specia, Lucia and Motta, Enrico (2007). Bridging the
gap between folksonomies and the semantic web: an experience report. In: The 4th
European Semantic Web Conference 2007 (ESWC 2007), 3-7 Jun 2007,
Innsbruck, Austria.
[23] C. Van Damme, M. Hepp , K. Siorpaes “Folksontology: An integrated approach for
turning folksonomies into ontologies,” In ESWC workshop. Bridging the Gap
between Semantic Web and Web 2.0 (2007), 2007.
[24] A. Passant, “Using ontologies to strengthen folksonomies and enrich information
retrieval in weblogs,” presented at the Proceedings of International Conference on
Weblogs, 2007.
[25] D. Torres, A. Diaz, H. Skaf-Molli, and P. Molli, “Semdrops: A Social Semantic
Tagging Approach for Emerging Semantic Data,” IEEE/WIC/ACM International
Conference on Web Intelligence (WI 2011), Aug. 2011.
[26] B. Huynh-Kim Bang, E. Dané, and M. Grandbastein, “Merging semantic and
participative approaches for organising teachers' documents,” presented at the In J.
Luca & E. Weippl (Eds.), Proceedings of World Conference on Educational
Multimedia, Hypermedia and Telecommunications 2008 (pp. 4959-4966).
Chesapeake, VA: AACE, Vienna, 2008.
[27] V. Tanasescu and O. Streibel, “Extreme tagging: Emergent semantics through the
tagging of tags,” Proceedings of the First International Workshop on Emergent
Semantics and Ontology Evolution, ESOE 2007, co-located with ISWC 2007 +
ASWC 2007, Busan, Korea, November 12th, 2007, 2007.