Linking Folksonomies to Knowledge Organization Systems

Abstract. This paper demonstrates enrichment of set-model folksonomies with hierarchical links and mappings to other knowledge organization systems. The process is exemplified with social tagging practice in Wikipedia and in Stack Exchange. The extended folksonomies are created by crowdsourcing tag names and descriptions to translate them to linked data in SKOS.

Keywords: social tagging, mapping, linked data, SKOS, crowdsourcing, digital libraries (cs.DL), information retrieval (cs.IR)

1 Introduction

With the rise of social software and web applications, social tagging has become a popular method to organize collections. Tagging is a process where keywords (tags) are manually assigned to a resource for retrieval. In contrast to traditional subject indexing, keywords are generally chosen freely by users and shared in a community. Many forms and applications of social tagging exist and are subject to research [4]. The outcome of activities in a tagging system is called a folksonomy. This paper first summarizes basic properties of social tagging with Stack Exchange and Wikipedia as two popular instances of set-model folksonomies. These systems are then compared to knowledge organization systems and the enrichment with hierarchical links and mappings to other systems via crowdsourcing is demonstrated.

2 Social Tagging and Folksonomies

No common definition of social tagging and folksonomies exists among authors and disciplines. As terms like social tagging, social indexing, and collaborative tagging are used interchangeably, tagging can be defined as manual indexing on the Web [14]. The specific type of a tagging system depends on several parameters [4,8,14]. In particular, tagging properties include:

Most research on tagging systems focuses on bag-model tag aggregation. This means that each resource can be tagged multiple times and every user can choose his individual set of tags to describe the resource. Folksonomies in bag-model tagging systems emerge as implicit consensus from large numbers of tagging events. Several approaches exist to derive folksonomies from tagging data by statistical analysis, including tag connectivity [1,5,7,10,11,12]. With Set-model tag aggregation in contrast there is only one common set of tags for each resource. Hence, the folksonomy is more directly given as snapshot of community consensus. Folksonomies expressed in set-model tagging systems can be defined as dynamic knowledge organization systems created by communities of distributed volunteers. Two popular instances are presented in the following with tags in Stack Exchange and categories in Wikipedia.

2.1 Tags in Stack Exchange

Stack Exchange is a growing network of question & answers communities with Stack Overflow as first and most prominent instance.¹ All content is licensed under Creative Commons Attribution Share Alike license (CC BY-SA) and accessible via a public API. Since Stack Overflow was launched in 2008, the Stack Exchange network has grown to almost 90 sites with 2 million users, 5 million questions, and 9 million answers (as of autum 2012). Since 2010 there is some academic research about Stack Exchange and the data that is provided by Stack Exchange sites.² Most of this research is focused on factors of success, quality and motivation and similar aspects of crowdsourcing. The tagging system of each community has not been analyzed yet. In Stack Exchange sites up to five tags are assigned to each question by its author. Reuse of existing tags is encouraged by typeahead suggestions and by limiting creation of new tags to experienced members of the community. Users with some reputation can also modify the tag-set of any question.³ Each tag can be defined with a short tag excerpt and a more detailed tag description, both editable in a wiki. Hierarchical links between tags are not supported on purpose.⁴ Figure 1 shows the extended info page of a tag with its tag excerpt and tag description.

2.2 Categories in Wikipedia

In Wikipedia articles are tagged by so called categories, which can be assigned and modified together with the normal content of an article.⁵ Categories are used for knowledge organization and for quality management, for instance to flag articles that lack references. Each category is described with a wiki page of its own. Category pages can be assigned to other categories, resulting in a directed graph of categories. The category system of Wikipedia is a thesaurus with similar structural and statistical properties like other social tagging and knowledge organization systems [16]. In addition to categories, most articles in Wikipedia can be used as concepts for knowledge organization. Wikipedia articles and categories translated to SKOS/RDF are provided by DBPedia project [2], including mappings from articles to authority files [15].⁶

3 Knowledge Organization Systems

Knowledge organization systems include systems such as classifications, taxonomies, thesauri, and authority files [6,13]. Each system defines a set of concepts that are used for the creation of metadata in digital libraries. Depending on context and community, knowledge organization systems are also known as controlled vocabularies, terminologies, and ontologies. An important topic in the research on networked knowledge organization systems (NKOS) is semantic interoperability of multiple systems via mappings and cross-concordances. or alignments. As defined by Mayr and Petras in the KoMoHe project [9], cross-concordances consist of manually created, directed relations between controlled terms of two knowledge organization systems. Mapping relations include equivalence, hierarchy, and association, possibly extended by a degree of confidence. To express and exchange mappings between knowledge organization systems, a common model of all connected systems is required. The most prominent model by now is the Resource Discovery Framework (RDF) in general and the Simple Knowledge Organization System (SKOS) in particular, covering the most common types of thesauri, authority files, and mappings. For instances of vocabularies and mappings published in SKOS/RDF see AGROVOC [3], TheSOZ [18], and the Library of Congress Subject Headings (LCSH). In SKOS/RDF each concept is identified by an URI and concepts are linked with a predefined set of RDF properties (table 1). Synonyms can be combined as multiple labels of one concept.

The relations skos:broader/broaderMatch and skos:narrower/narrowMatch are inverse respectively, the other relations are symmetric, and skos:exactMatch is transitive. More elaborated models of cross-concordances allow for non-symmetrical and single-to-multiple mapping relations. [9,12]. The SKOSified terminologies presented in this paper make use of th relations skos:broader, skos:narrower, skos:related, and skos:closeMatch (to avoid transitive mappings). Hierarchical mappings will be added later.

Relations between concepts in the SKOS model
concept relation	mapping relation	purpose
skos:broader	skos:broadMatch	direct hierarchical link (up)
skos:narrower	skos:narrowMatch	direct hierarchical link (down)
skos:related	skos:relatedMatch	associative link
	skos:closeMatch	equivalence link with low confidence
	skos:exactMatch	equivalence link with high confidence

4 From Folksonomies to Knowledge Organization Systems

A set-model based folksonomy is continuously modified and extended by members of a community. The volunteers make use of tagging not to create a reusable folksonomy but as tool for knowledge organization within their project. Because of the open and dynamic nature of the projects, nobody is responsible for the full tagging terminology. This makes centralized approaches to enrich the folksonomy difficult. For this reason additional mapping and linking can best be managed within the tagging system. If enrichment is also done by the community, it can be crowdsourced together with the folkosonomy. Two methods of seamless integration are presented below.

4.1 Links from tag names

The first method to link a folksonomy with a knowledge organization system is used at the Stack Exchange site about theoretical computer science.⁷ In this community some tags follow the syntax “xx.name” where “xx” is part of a notation from the classification of the Computing Research Repository (CoRR)⁸ and “name” is a descriptive name. For instance the tag “lo.logic” refers to the category “Logic in Computer Science” with CoRR notation “cs.LO”. Some CoRR categories have no tag at cstheory.stackexchange and for some categories multiple tags exists. Based on this tag naming rules, a formal SKOS mapping can be derived with 1-to-1 close/exact matches and 1-to-n narrower/broader matches:

To illustrate the use of this mapping, a simple statistical analysis was conducted. The total number of computer science papers archived at arXiv.org in 2011 for each CoRR category was compared to the number of question tagged with corresponding tags. Appendic A lists the 16 CoRR categories with at least 10 related questions and the number of papers per question. One can see that there are more research papers in artificial intelligence, computer vision and pattern recognition, and information theory compared to more questions in computational complexity, algorithms and data structures, computational geometry, and programming languages.

4.2 Links from tag descriptions

Both Stack Exchange and Wikipedia have a wiki page for each tag, which can be edited independently from the act of tagging single questions or articles. This form of tag management can be used to express more elaborated types of links between the folksonomy and other knowledge organization systems. Participation in this enrichment, however, is lower than tagging activity because tag descriptions are less visible members of the communities. Figure 1 shows the tag description of tag ils.⁹

The wiki contains HTML links to other tags and links that make use of concepts from other knowledge organization systems (Wikipedia, LCSH, JITA classification, and GND authority file) to get related resources. These links can be harvested via Stack Exchange API and translated to semantic relationships in SKOS. The translation between HTML links in the tag description and URIs in the linked system must be configured for each. For instance a link to Wikipedia is translated to DBPedia and a link to a Worldcat search by LCSH is translated to an URI at http://id.loc.gov. This results in the following concept in SKOS/RDF:¹⁰

A similar method has been applied experimentally in German Wikipedia to link categories with other knowledge organization systems. Figure 2 shows the category description of category “Hörspiel” (radio play). An infobox is used to show links to corresponding concepts in Regensburger Verbundklassifikation (RVK), Dewey Decimal System (DDC), and GND authority file.

Translation of these links to mappings in SKOS is based on the template syntax of MediaWiki. If multiple links are specified to the same system, as RVK in the example, the relation skos:narrowMatch is used instead of skos:closeMatch. The translation results in the following RDF statements (hierarchical and associative relations between categories are omitted because they are already included in DBPedia):

Adoption of category descriptions enriched with mappings in Wikipedia is still low because category pages are less visible to Wikipedia users and because creation of a mapping requires knowledge of the linked knowledge organization system. A special mapping tool may boost, such as the tool that was used to match biographic articles in German Wikipedia and GND authority files [15].

5 Summary and Discussion

Two instances of folksonomies with set-model tag aggregation have been presented with tags in Stack Exchange and categories in Wikipedia. In contrast to bag-model tagging systems, enriched folksonomies cannot be calculated but one must explicitly express links and mappings to other knowledge organization systems. Tag names and tag descriptions can be used to express these additional connections. Curation of links and mappings by the social tagging community depends on visibility (tagging support) and ease of tagging. Simple equivalence links, which make up 45% of typical mapping relations [9] are easier to manage. These mapping can also be provided in simplified form, such as BEACON files which are also used to map GND authority records, Wikipedia and other resources [17]. It is shown how links from tag names and tag descriptions can be harvested and transformed to concept schemes in SKOS. The resulting knowledge organization systems can be used for retrieval, to find related resources, and for bibliometric analysis as exemplified in Table 2.

Popular cstheory tags mapped to CoRR categories
category	papers	questions	relation	tags
cs.AI	788	42	18.76	ai.artificial-intel
cs.CC	421	931	0.45	cc.complexity-theory
cs.CG	225	133	1.69	cg.comp-geom
cs.CR	485	143	3.39	cr.crypto-security
cs.CV	384	11	34.91	cv.computer-vision
cs.DB	244	29	8.41	db.databases
cs.DC	450	93	4.84	dc.parallel-comp, dc.distributed-comp
cs.DS	800	915	0.87	ds.algorithms, ds.data-structures
cs.FL	194	150	1.29	fl.formal-languages
cs.GT	324	34	9.53	gt.game-theory
cs.IT	1692	41	41.27	it.information-theory
cs.LG	464	53	8.75	lg.learning
cs.LO	567	151	3.75	lg.logic
cs.NA	137	15	9.13	na.numerical-analysis
cs.NE	150	23	6.52	ne.neural-evol
cs.PL	242	122	1.98	pl.programming-languages

6 References

[1] Barla, M., Bieliková, M.: On Deriving Tagsonomies: Keyword Relations Coming from Crowd. In: ICCCI 2009, pp. 309-320, Springer (2009)
[2] Christian Bizer, C., Lehmann, J., Kobilarov, G., Auer, S., Becker, C., Cyganiak, R., Hellmann, S.: DBpedia – A Crystallization Point for the Web of Data. Journal of Web Semantics: Science, Services and Agents on the World Wide Web. 7, 154–165 (2009)
[3] Caracciolo, C., Morshed, A., Stellato, A., Johannsen, G., Jaques, Y., Keizer, J.: Thesaurus Maintenance, Alignment and Publication as Linked Data: The AGROVOC Use Case. In: García-Barriocanal, E., Cebeci, Z., Okur, M., Öztürk, A. (eds.) Metadata and Semantic Research, pp. 489-499, Springer (2011)
[4] Gupta, M., Li, R., Yin, Z., Han, J.: An Overview of Social Tagging and Applications. In: Aggarwal, C. (eds.) Social Network Data Analytics, pp. 447-497, Springer (2011)
[5] Heymann, P., Garcia-Molina, H.: Collaborative creation of communal hierarchical taxonomies in social tagging systems. Technical report, Stanford University (2006)
[6] Hodge, G.: Systems of knowledge organization for digital libraries (2000)
[7] Lin, H., Davis, J., Zhou, Y.: An integrated approach to extracting ontological structures from folksonomies. In: ESWC 2009, pp. 654-668 (2009)
[8] Marlow, C., Davis, M., Boyd, D.: HT06, tagging paper, taxonomy, Flickr, academic article, ToRead. In: Proceedings of Hypertext 2006 (2006)
[9] Mayr, P., Petras, V.: Cross-concordances: Terminology mapping and its effectiveness for information retrieval. In: 74th IFLA World Library and Information Congress (2008)
[10] Moosavi, A., Li, T., Lakshmanan, L., Pottinger, R.: ONTECTAS: Bridging the Gap Between Collaborative Tagging Systems and Structured Data. In: CAiSE conference, pp. 436-451 (2011)
[11] Solskinnsbakk, G., Gulla, J.A.: A Hybrid Approach to Constructing Tag Hierarchies. In: OTM Conferences, pp. 975-982 (2010)
[12] Specia, L., Motta, E.: Integrating folksonomies with the semantic web. In: Proceedings of the ESWC 2007, pp. 624-639, Springer (2007)
[13] Tudhope, D., Koch, T.: New Applications of Knowledge Organization Systems: introduction to a special issue. Journal of Digital Information, 4 (2006)
[14] Voss, J.: Tagging, Folksonomy & Co-Renaissance of Manual Indexing? (2007)
[15] Voss, J.: Wikipedia als Teil einer freien bibliothekarischen Informationsinfra- struktur, In: Daniela Lülfing (ed.) 95. Deutscher Bibliothekartag, pp. 63-74, Klostermann (2007)
[16] Voss, J.: Collaborative thesaurus tagging the Wikipedia way (2006)
[17] Voss, J., Schindler, M., Thiele, C.: Link server aggregation with BEACON. In: International Symposium for Information Science (2011)
[18] Zapilko, B., Schaible, J., Mayr, P., Mathiak, B.: TheSoz: A SKOS Representation of the Thesaurus for the Social Sciences. Journal of Semantic Web (2012, to appear)

See http://stackexchange.com and http://stackoverflow.com.↩
See http://meta.stackoverflow.com/questions/134495 for a bibliography.↩
Editing rights in Stack Exchange are controlled by an elaborated system of reputation points. In Stack Overflow 500 points are required for retagging and 1500 for creating tags. In beta sites levels are 200 and 150 respectively.↩
See http://meta.stackoverflow.com/questions/tagged/tag-hierarchy for discussion of the decision against tag (mono)hierarchies.↩
See http://en.wikipedia.org/wiki/Wikipedia:Categorization.↩
Available at http://dbpedia.org.↩
http://cstheory.stackexchange.com↩
The Computing Research Repository (http://arxiv.org/corr) is part of the arXiv repository.↩
Available at http://libraries.stackexchange.com/tags/ils/info.↩
Scripts to download/transform links are available at https://github.com/nichtich/se2skos.↩