Author: George Anadiotis
Date: 2006-10-11
In this document, we will give an overview of user created metadata in the form of tags (user-provided keywords that are assigned to resources) and try to formulate some related questions, as well as outline answers to these questions:
Metadata is often characterized as “data about data.” Metadata is information, often highly structured, about documents that is used to support specific functions; these functions are usually to facilitate some organization and access of information. Different classifications of metadata can be used, depending on the classification criteria.
If we classify metadata in terms of what they describe (metadata types), then we distinguish between administrative, structural, and descriptive metadata [17].
In terms of their source, metadata can be classified as professional, author and user created.
Metadata that are created by dedicated professionals usually employ some of the elaborate rules and schemes developed by library and information science for cataloging, categorization and classification. While professionally created metadata are often considered of high quality, it is costly in terms of time and effort to produce.
An alternative is author created metadata: original creators of content provide metadata along with their creations. While author created metadata may help with the scalability problems in comparison to professional metadata, both approaches share a basic problem: the intended and unintended eventual users of the information are disconnected from the process. In both cases, end users have to be familiar with the terminology used in order to be able to effectively seek information.
User created metadata is a third approach, which lets the metadata provision task in the hands of end-users. User created metadata are considered to be on the opposite end of the spectrum compared to professional created metadata: cheap to acquire, but low in quality.
One of the currently most prominent efforts towards standardization and semantic use of metadata on the WWW is the Semantic Web. The Semantic Web initiative aims to provide technologies and tools that support the addition of a meaning layer above the conventional web. Through a technology stack that includes the Resource Description Framework (RDF) and Web Ontology Language (OWL), the goal of the semantic web is the creation of machine-readable descriptions that allow content managers to add meaning to content.
The role the Semantic Web standards are expected to play is analogous to the role the WWW standards played in the early stages of its deployment: to promote interoperability and allow the focus to be shifted towards applications of the technology.
Tags are user-provided keywords that are assigned to resources. They are mostly used in order to support subsequent retrieval. In contrast to the formal semantics provided by the Semantic Web standards, tags have no semantic relations whatsoever, including a lack of hierarchy; tags are just flat collections of keywords.
In itself, this idea is not new or revolutionary: keyword-based retrieval has been around for a while. There are however two factors that have made tagging attract attention and differentiate it from traditional keyword assignement: low-cost applicability and collaborative tagging.
One of the reasons that have been hindering the widespread utilization of Semantic Web standards for metadata provision is that, as things stand now, it is rather difficult to use Semantic Web standards without having a certain level of comprehension of formal logic principles, as well as familiarity with specialized tools that support users in the annotation process.
Tagging on the other hand lowers the barrier of metadata annotation, since it requires minimal effort on behalf of annotators: there are no special tools or complex interface that the user needs to get familiar with, and no deep understanding of logic principles or formal semantics required – just some standard technical expertise.
While user-provided metadata in general are often referred to as being of low quality, there is evidence that this is not necessarily so. In the user study conducted in [6], untrained users were asked to provide metadata (tags) for museum exhibits. The results showed that 77% of the user-provided terms were determined to be valid.
The main idea behind collaborative tagging is simple: collaborative tagging platforms (CTPs, or alternatively, distributed classification systems - DCSs [4]) provide the technical means, usually via some sort of web-based interface, that support users in tagging resources. What is the important aspect of this is that they aggregate collections of tags that an individual uses, or his tag vocabulary, called a personomy [13], into what has been termed a folksonomy: a collection of all personomies [18, 21].
Some of the most popular CTPs are Delicious (bookmarks), Flickr (images), Last.fm (music), YouTube (video), Connotea (bibliographic information), steve.museum (museum items) and Technorati (blogging). Using these platforms is free, although in some cases users can opt for more advanced features by getting an upgraded account, for which they have to pay. The most prominent among them are Delicious and Flickr, for which some quantitative user studies are available [1,2]. These user studies document a phenomenal growth, that indicates that in real-life tagging is a very viable solution for annotating any type of resource.
As there is a multitude of CTPs, there are also different approaches in terms of design and characteristics, which may be used to derive a taxonomy of CTPs. These design and characteristics properties, as identified in [5], are:
There are 3 aspects of collaborative tagging that set it apart from traditional keyword assignment: volume, interaction and navigability.
Volume is important because the more users annotate a resource, the more complete the annotations get. In [6], it was discovered that 88% of the terms supplied by users were not supplied by professional annotators. Offering more valid terms for resources is clearly benefitial, as it creates alternative points of access.
Or, as stated in [33], 'with a multiplicity of points of view the question isn't "Is everyone tagging any given link 'correctly'", but rather "Is anyone tagging it the way I do?" As long as at least one other person tags something they way you would, you'll find it'.
CTPs are also referred to as'social software': software that fosters social interaction. This makes CTPs display an emergent semantics [34] behaviour: there is no a priori semantic commitment, but rather negotiations through interaction that make global agreements emerge by aggregating local agreements. This aspect of collaborative tagging has been suggested in [13] and verified in [35], and offers an alternative to traditional ontological commitment.
A very interesting feature that CTPs offer is navigation: by selecting a tag T, the user has the option to view other items that have also been tagged with T, be it by the same user or by others as well. One side effect that stems from the navigability of tags is the creation of what has been termed 'feral hypertext' [8] : arbitrary links between documents, that the author did not foresee. These links are instead added by the document consumers.
There are however certain limitations to the unstructured tagging approach. These may be demonstrated by a motivating example:
Let's suppose that user Mary has an account on platform S1, that specializes in images. Mary has been using S1 for a while, so she has progressively built a large image collection, as well as a rich vocabulary of tags (personomy).
Another user, Sylvia, who is Mary's friend, is using a different platform, S2, to annotate her images. At some point, Mary and Sylvia attended the same event, and each one took some pictures with her own camera. As each user has her reasons for choosing a preferred platform, none of them would like to change. They would like however to be able to link to each other's annotated pictures, where applicable: it can be expected that since the pictures were taken at the same time and place, some of them may be annotated in similar way (same tags), even by different annotators. So they may (within the boundaries of word ambiguity) be about the same topic. This is currently not possible however, as each platform uses ad-hoc solutions and only provides tag navigation within its own boundaries: there is no standardization/syntactic interoperability.
In the course of time Mary also becomes interested in video and starts shooting some of her own. As she initially wasn't planning on making these videos available to anyone else, she stores them locally in her PC, using the file system to organise them into folders. This hierarchical organisation scheme already bears some semantic information [3].
At some point, Mary decides to switch from storing her video collection locally, to using another CTP, S3, that specializes in video. This switch however is not flawless: the semantics captured in the folder hierarchy (hypernym and hyponym) will be lost. In most of the cases, existing hierarchical structures utilized by users are simply not taken into account: all the information that is encoded e.g. as folder names is discarded, and has to be re-entered by the user, despite the fact that it could be used as tags for the items the folders contain.
Another alternative for 'personomy bootstrapping' in the S3 platform would be to reuse the personomy Mary has already built in S1: while some of the tags may not be appropriate, as they may represent one-off ('29-08-06') or photography-specific ('CameraXYZ') use, others might as well be reused across modalities/domains, in case they represent high-level concepts ('holidays'). So if Mary has both video and photographic material of some event, and since she has already created a personomy on S1, she would naturally like to be able to reuse it (partially, perhaps) on S2 as well. This emphasizes, from a different viewpoint, the lack of standardization/syntactic interoperability.
Let us now imagine that a third person, John, is also using S1. In fact, he is also using a tag that Mary is using, T1, but in a totally different context: as John's preferred language is different than Mary's, they give T1 a different meaning. Nevertheless, navigating S1 would present the items Mary and John have tagged with T1 in the same set, as there is no possibility to semantically correlate tags to other tags, by denoting relations such as synonym and antonym. There is the implicit assumption that two tags that match on a string level have the same meaning as well.
After using any of the systems mentioned above for a while, Mary recognizes that assigning tags appropriately to new content, e.g. images or videos, becomes difficult as the number of tags in her personomy has grown. Furthermore, for many tags she does not know what is the distinction between them. Some systems allow to group tags in bundles which effectively adds one level of hierarchy to her personomy, however, only pushes the problem one step ahead. In the end Mary often feels overwhelmed by the size of her personomy so that using it for organizing and finding her content becomes a burden for her.
Finally, Mary realizes that the navigational/serendipitous aspect of the aforementioned systems differentiates them from the way navigation is usually implemented on the WWW. While on the WWW it is the author of a document that determines the document's links, in CTPs users can create navigation links between resources by annotating them. By using any of the aforementioned systems she is able to annotate items, see other users' annotations for items, and also navigate among items that have similar annotations, so she would like to be able to use this functionality unrestricted and integrated into her everyday WWW browsing.
Currently however this is not possible. If, for example, she wants to annotate an image she found while browsing on the web, she would have the following options:
This requires additional effort (downloading/uploading), produces multiple copies of the content and detaches the annotation from its original target, thus making it non-reachable for users who view the original image.
Although this option does not introduce the intermediate download/upload step and does not produce multiple copies of the content, it still detaches the annotation from its original target. In addition, searching for annotations is quite awkward, as it requires manual copy-pasting of the URL to the tagging platform's search engine.
Most importantly, both solutions are proprietary and based on privately run centralized repositories, since there is no way for a user that does not have (write) access to some metadata repository / publishing platform to publish resources and add annotations about them. There is also no universally accepted specification for tags and no clearly defined source or protocol to look them up.
The above scenarios demonstrate some currently well-known limitations of tag-based systems:
Before embarking on a description of possible solutions for these issues, we try to outline some requirements:
We will henceforth reference these requirements as R1, R2, R3 etc.
So, let's see some possible solutions to the issues spotted above:
When it comes to interoperability, standards-based solutions have repeatedly proven successful in enabling to bridge different systems. This could also be the case here, as a standard for expressing personomies and folksonomies would enable interoperability across platforms. On the other hand, use of a standard should not enforce changes in the way tags are handled internally by each system - it simply aims to function as a bridge between different systems. The question is then, what standard?
We may be able to answer this question if we consider a personomy as a concept scheme: tags used by an individual express his or her expertise, interests and vocabulary, thus constituting the individual's own concept scheme. A recent W3C standard that has been designed specifically to express the basic structure and content of concept schemes is SKOS Core [9].The SKOS Core Vocabulary is an application of the Resource Description Framework (RDF), that can be used to express a concept scheme as an RDF graph. Using RDF allows data to be linked to and/or merged with other RDF data by semantic web applications.
Expressing personomies and folksonomies using SKOS is a good match for covering R1 and R2: not only does it enable expression of personomies in a standard format that fits semantically, but also allows mixing personomies with existing Semantic Web ontologies, as will be shown in the following example. This lays the foundation for providing solutions for R3 and R5, as will be detailed in sections 'Personomy management and usability' and 'Semantic enrichment of personomies' respectively.
There is already a publicly available SKOS-based tagging ontology that can be used to build on [31].
Currently, all that users of a tagging platform can do with their personomies is view them, either as a list or as a cloud. There is no option to export/import personomies, so if user XYZ of platform tagplatform.com wants to get a list of his tags, there is really nothing else he can do except to manually create this by copy-pasting into any proprietary text-based format. If we make the simplistic assumption that userXYZ is only using Tag1 and Tag2, then one such text-based format could be:
Tag1 Tag2
This minimal text format gives no further information whatsoever about the user's personomy. If we want to encode such information, we would have to come up with a proprietary text-based format that could be used to exchange personomy information. It makes much more sense to use an existing appropriate standard like OWL/SKOS to express personomies.
Another issue is how to mint URIs that refer to a personomy. Since we refer to personal use, owning a domain cannot be taken for granted. Therefore we could either make up an imaginary domain for each user, and use it to come up with a URL, or, alternatively, use the tag URI scheme [36] that allows minting URIs similar to URNs (also non dereferenceable) based on personal data, e.g. email address. In this example we choose the 2nd option.
<?xml version="1.0" encoding="UTF-8"?> <!DOCTYPE rdf:RDF [] > <rdf:RDF xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:dcterms="http://purl.org/dc/terms/" xmlns:owl="http://www.w3.org/2002/07/owl#" xmlns:skos="http://www.w3.org/2004/02/skos/core#" xmlns:tag="http://www.holygoat.co.uk/owl/redwood/0.1/tags/"> <skos:ConceptScheme rdf:about="tag:userXYZ@mydomain.com,2006:mytags"> <skos:hasTopConcept rdf:resource="Tag1"/> <skos:hasTopConcept rdf:resource="Tag2"/> <dcterms:issued>2005006T120000+0000</dcterms:issued> <dcterms:modified>20060818T120000+0000</dcterms:modified> <dc:title>My personomy</dc:title> <dc:rights>Copyleft</dc:rights> </skos:ConceptScheme> <Tag1 rdf:type="&tag;Tag"> <skos:inScheme rdf:resource="tag:userXYZ@mydomain.com,2006:mytags"/> <skos:prefLabel xml:lang="en-GB">Tag1</skos:prefLabel> <skos:altLabel xml:lang="en-US">xxx</skos:altLabel> <skos:definition xml:lang="en-GB">Free-text description of Tag1</skos:definition> <skos:changeNote xml:lang="en-GB">renamed - old name was ttt1</skos:changeNote> </Tag1> <Tag2 rdf:type="&tag;Tag"> <skos:inScheme rdf:resource="tag:userXYZ@mydomain.com,2006:mytags"/> <skos:prefLabel xml:lang="en-GB">Tag2</skos:prefLabel> <skos:altLabel xml:lang="en-US">yyy</skos:altLabel> <skos:definition xml:lang="en-GB">Free-text description of Tag2</skos:definition> <skos:changeNote xml:lang="en-GB">none</skos:changeNote> </Tag2> </rdf:RDF>
Since ease of use is a factor that largely contributes to the adoption of tagging, further usability enhancements will encourage even more users to participate in the annotation process, as well as decrease the time needed to annotate. Thus, by covering R4 it should be possible to increase both the quantity and the quality of the annotations. One way to achieve this is to assist users in the annotation process, by suggesting candidate tags for resources.
Currently, the only CTP that provides some sort of bootstrapping facility is Delicious. Delicious supports an 'import' functionality, that exploits bookmark folder structure to extract tags that are associated to bookmarks added in this batch import mode: a bookmark in the “recipes” folder inside a “household” folder will be tagged both “household” and “recipes.” Even in this case however, the hierarchical organisational structure of the folders is lost, as tags are inherently unstructured.
One possible way to deal with the bootstrapping issue could be the development of a tag extraction mechanism. This mechanism should try to extract tags from all possible sources of metadata, in order to suggest to the user a wide range of possible tags to assign to a resource. Tags can be mined from hierarchical structures such as filesystem folders, bookmarks and e-mail organisation associated to the user, but also from the items stored in these structures as well as from the way other users tag.
As pointed out in [3], hierarchical organisation schemes used for classification in filesystems already bear some semantic information in the form of folder names and hierarchy, although the semantics are not always clear. One solution that already applies a technique that utilizes hierarchical structure for tag extraction is TagFS [12], a filesystem with tagging support which aims to integrate the tagging paradigm with local applications. What TagFS effectively achieves is the extraction of tags from a hierarchical filesystem organisation; these tags can subsequently be used to classify and retrieve files using multiple hierarchical structures, and are also stored in a local RDF repository (hierarchical scheme tag extraction). It is generally not possible to know the type of metadata that are extracted using this method, since folder names may represent any kind of arbitrary classification.
Another way to harvest tags would be to take advantage of metadata already present (mime-type specific metadata tag extraction). This may include for example EXIF metadata for images, ID3 metadata for audio tracks, meta information for HTML and so on. This requires the utilization of a mime-type specific mechanism that will leverage existing metadata formats. This type of metadata usually refers to administrative aspects.
File content itself can also be used to mine tags (content-based tag extraction). In the same manner that document-specific techniques are utilized for HTML in [7], multimedia-specific techniques could be utilized for audio or images. For example, suggesting tags such as 'slow', 'mid-tempo' 'fast' for audio tracks, according to their (automatically extracted) BPM , or similarly suggesting tags such as 'blue', 'red' etc according to the dominant colour extracted for images. The tags extracted using such techniques will most likely be classified as content metadata.
Finally, another technique that is already used e.g. by Delicious, is to suggest tags for items based on existing annotations for the same item added by other users (suggestive tagging). This obviously applies only to self-tagging platforms such as Delicious (since in a free-for-all platform it makes no sense to re-enter the same tag for a resource), and has the side-effect that it makes the vocabulary converge faster [22].
Building on a SKOS-based representation of personomies, there is the possibility to leverage semantically rich relations as enabled by standard OWL properties, as well as SKOS properties and custom properties. This covers R5.
As far as the hypernym and hyponym distinction is concerned, using SKOS allows us to express something that is currently not expressible using tags: hierarchical relations.
Delicious has taken a first step towards bringing some sort of hierarchy in personomies, by introducing tag bundles [11]. Tag bundles represent simple tag sets, only used for display/organisation purposes: tag bundles themselves are just labels, since they cannot be used as tags, and there is also no nesting of tag bundles. They can be interpreted either as a one-level hierarchy of tags, or as a set of related tags.
There is also recent work [32] that applies statistical techniques to induce hierarchy in a folksonomy. It does so by examining co-ocurrence of tags in resources and extracting subsumption relations thereof. Although this might prove useful, the focus of that work is slightly different, as it aims in inducing an ontology to be used internally in the tagging platform under question (in this case, Flickr).
Regarding the synonym and antonym relations between tags, again, utilizing Semantic Web standards to express personomies enables us to leverage semantic relations explicitly designed to model such distinctions, such as 'owl:sameAs' and 'owl:differentFrom'. Some similar functionality has been proposed in [15], although that proposition only covers mapping of tags and not disambiguation.
Keeping the previous example as a reference, let's see how we can use existing language constructs to state additional semantic relationships. So, if we want to state that Tag1 is a generalisation of Tag2, we can do so by adding a pair of properties, let's call them subtag and supertag, that extend skos:broader and skos:narrower respectively, to the previous example:
<Tag1 rdf:type="&tag;Tag"> <supertag rdf:resource="#Tag2"/> ..... </Tag1> <Tag2 rdf:type="&tag;Tag"> <subtag rdf:resource="#Tag1"/> ..... </Tag2>
If we furthermore want to disambiguate between tags belonging either to the same or to different personomies, we can use the owl:sameAs and owl:differentFrom statements to achieve this. So, if want to state that Tag1 has the same meaning as some other TagA belonging to user1's personomy, but has a different meaning from Tag1 as used in user2's personomy, then the example becomes:
<Tag1 rdf:type="&tag;Tag"> <tag:equivalentTag rdf:resource="tag:user1@mydomain.com,2006:mytags#TagA"/> <owl:differentFrom rdf:resource="tag:user2@mydomain.com,2006:mytags#Tag1"/> ..... </Tag1>
Currently, personomy management is fragmented among different tagging platforms, so there is no single point of control for the user. In addition, interoperability / bootstrapping / semantic enhancement solutions should be handled transparently. The goal is to build on ease of use by assisting the user in performing personomy administration tasks (R7):
These services, which correspond to issues/solutions mentioned, can be bundled together as a 'personomy manager'. Each function it performs can be structured so that it operates more or less independently from the rest, thus allowing this 'personomy manager' to be in the form of an application/service/website that could be also reused in a modular fashion.
One way to circumvent the fact that currently, only resources within the boundary of each specific tagging platform can be associated via annotation to other similarly tagged resources is the use of URLs as tags, as documented in [16]: in addition to using keywords as tags, some users have also started using URLs as tags. Tags of this kind are a special case, as in contrast to 'normal' tags they provide links to documents that are ouside the scope of a specific tagging platform. Still, this link can only be uni-directional: the point of origin will always be a resource within the boundaries of a specific platform, therefore modality limitations apply. Tagging platforms focused on bookmark sharing (BSTP) could be considered the only exception to this rule, as in this case resources are also URLs.
However, there are still limitations in finding links created through user annotation for arbitrary resources: one has to manually switch to the BSTP for looking up user-created links for a specific URL, instead of being available in addition to author-created ones. Furthermore, even by resorting to the BSTP, there are limitations in the search functionality provided for URLs.
More generally, existing tagging platforms apply a centralized collaborative tagging approach. On the other hand, a solution such as the aforementioned personomy manager represents a change of perspective towards the individual user and focuses on ease of use, but lacks the collaborative aspect: it provides a single user unified view, but no interaction between personomies. It would then seem reasonable that the next step in this approach would be to cater for interaction between individual personomy managers (R8).
This interaction could be either mediated or direct: mediated interaction would mean that personomy managers do not communicate directly, but rather through separate CTPs or a dedicated platform that enables personomy manager interaction. Both solutions however are not ideal, as the former assumes adoption of the personomy manager standards by every CTP, while the second one does not deal with the dependency on a centralized solution issue.
Direct communication seems better suited for a user-centered approach, since it does not have any external dependencies: user tags and annotations may come from and be kept in sync with one or more existing CTPs, but it should be also possible to store them locally (in a way similar to what TagFS does). So, in that respect, it seems like the adoption of a peer-to-peer communication protocol for interaction between personomy managers would be a good match.
A requirement for the realization of the direct interaction scenario is the utilization of a dereferenceable protocol for communication. One approach for this could be the adoption of http - however, the main technical issue in that case would be the handling of dynamic IPs, which are used by the greater part of internet connections. A solution for this issue based on peer-to-peer protocols has already been proposed in [28], in which the peer-to-peer network itself is used to store the permanent identity of nodes.
By allowing each user-hosted personomy to become dereferenceable through peer-to-peer we promote detachment of personomies from centralized solutions. In addition, this also enables the automated discovery of user annotations while browsing the WWW, by querying the peer-to-peer network. This effectively constitutes a unified browsing experience, in which author-created and user-created content and metadata are intertwined (R9).
Some existing systems that adopt similar approaches (arbitrary resource annotation) are [7], [27], [29] and [30]. In [7] the focus is on a browsing environment that unifies the tagging navigation approach and peer-to-peer resource sharing , but it is not based on any standards and the approach to tagging is completely unstructured. [29] utilizes a tagging ontology that is populated in an RDF store and also focuses on the collaborative aspect of tagging by adopting the peer-to-peer paradigm for annotation exchange. However, it does not emphasize on the bootstrapping and semantic enhancement issues. [30] allows adding arbitrary annotations to web resources in the form of application specific objects (bookmarks and topics) and utilizes a centralized architecture. Finally, [27] explores the idea of decentralized user-provided annotation, allowing users to instantly add annotations on web resources and then share them through a peer-to-peer network, but is also not based on any standards.
We have presented some background as well as the current state of affairs on user-provided metadata in the form of collaborative tagging, as well as known issues pertaining this area. We have outlined requirements for an improved solution.
[1] HitWise Intelligence: Del.icio.us Traffic More Than Doubled Since January
http://weblogs.hitwise.com/leeann-prescott/2006/08/delicious_traffic_more_than_do.html
[2] Nielsen//NetRatings: USER-GENERATED CONTENT DRIVES HALF OF U.S. TOP 10 FASTEST
GROWING WEB BRANDS, ACCORDING TO NIELSEN//NETRATINGS
http://www.nielsen-netratings.com/pr/PR_060810.PDF
[3] Jones, W., Phuwanartnurak, A. J., Gill, R., and Bruce, H. 2005. Don't take my folders away!: organizing personal information to get things done. In CHI '05 Extended Abstracts on Human Factors in Computing Systems (Portland, OR, USA, April 02 - 07, 2005). CHI '05. ACM Press, New York, NY, 1505-1508. DOI= http://doi.acm.org/10.1145/1056808.1056952
[4] Ulises Ali Mejias, 2005: Tag literacy
http://ideant.typepad.com/ideant/2005/04/tag_literacy.html
[5] MARLOW C., NAAMAN M., BOYD D., DAVIS M. (2006), "Position paper, Tagging, Taxonomy, Flickr, Article, ToRead?", paper for the Collaborative Web Tagging Workshop at WWW2006, Edinburgh, 22 may 2006
http://www.rawsugar.com/www2006/29.pdf
[6] J. Trant (2006), "Exploring the potential for social tagging and folksonomy in art museums: proof of concept", New Review of Hypermedia and Multimedia, 2006
[7] J. Fokker, J. Pouwelse, W. Buntine, "Tag-Based Navigation for Peer-to-Peer Wikipedia", paper for the Collaborative Web Tagging Workshop at WWW2006, Edinburgh, 22 may 2006
http://www.rawsugar.com/www2006/9.pdf
[8] Walker, J. 2005. Feral hypertext: when hypertext literature escapes control. In Proceedings of the Sixteenth ACM Conference on Hypertext and Hypermedia (Salzburg, Austria, September 06 - 09, 2005). HYPERTEXT '05. ACM Press, New York, NY, 46-53. DOI= http://doi.acm.org/10.1145/1083356.1083366
[9] SKOS Core: http://www.w3.org/2004/02/skos/core/
[10] W3C Multimedia Semantics Incubator Group: http://www.w3.org/2005/Incubator/mmsem/
[11] del.icio.us bundles: http://del.icio.us/help/bundles
[12] Bloehdorn, S., Görlitz, O., Schenk, S., Völkel, M., "TagFS - Tag Semantics for Hierarchical File Systems". In Proceedings of the 6th International Conference on Knowledge Management (I-KNOW 06), Graz, Austria, September 6-8, 2006. September 2006.
http://www.aifb.uni-karlsruhe.de/WBS/sbl/publications/2006-09-iknow2006-tagfs.pdf
[13] Hotho, A., Jaschke, R., Schmitz, C., Stumme, G. (2006): Information Retrieval in Folksonomies: Search and Ranking. 3rd European Semantic Web Conference, 11-14 June 2006, Budva, Montenegro
http://www.kde.cs.uni-kassel.de/hotho/pub/2006/seach2006hotho_eswc.pdf
[14] Ahlstrom, V. (2005). A comparison of subject-based classification strategies for enhanced usability . Proceedings of the Human Factors and Ergonomics Society Annual Meeting, 49, 1439-1443.
http://acb220.tc.faa.gov/papers/ahlstrom_2005.pdf
[15] I. Ohmukai, M. Hamasaki, and H. Takeda: A Proposal of Community-based Folksonomy with RDF Metadata. Workshop on End User Semantic Web Interaction, Colocated with the Fourth International Semantic Web Conference (ISWC2005), 2005
http://www.ifi.unizh.ch/ddis/fileadmin/events/iswc2005ws/CameraReady/Ohmukai_Folkosonomy_021.pdf
[16] M. Guy and E. Tonkin. Folksonomies: Tidying up tags? D-Lib Magazine, 12(1), January 2006
http://www.dlib.org/dlib/january06/guy/01guy.html
[17] Taylor, Arlene G. The Organization of Information. Westport: Libraries Unlimited 2004
[18] Mathes, A., Folksonomies - Cooperative Classification and Communication Through Shared Metadata. Computer Mediated Communication - LIS590CMC, Graduate School of Library and Information Science, University of Illinois Urbana-Champaign, 2004
http://www.adammathes.com/academic/computer-mediated-communication/folksonomies.html
[19] Porter, J., The Del.icio.us Lesson
http://bokardo.com/archives/the-delicious-lesson/
[20] Menchen-Trevino, E., Feedback, Motivation & Sociality in an Online Tagging Service - de.licio.us user survey
http://www.erickaakcire.net/delicious/pilot
[21] Smith, G. “Atomiq: Folksonomy: social classification.” Aug 3, 2004
http://atomiq.org/archives/2004/08/folksonomy_social_classification.html
[22] Scott Golder and Bernardo A. Huberman. (2006). "Usage Patterns of Collaborative Tagging Systems." Journal of Information Science, 32(2). 198-208.
http://www.hpl.hp.com/research/idl/papers/tags/tags.pdf
[23] Adar, E., Huberman, B. Free riding on Gnutella. Technical report, Xerox PARC, 10 Aug. 2000.
[24] J. Shneidman and D. Parkes. Rationality and self-interest in peer to peer networks. Int. Workshop on Peer-to-Peer Systems (IPTPS), 2003.
[25] Feigenbaum, J. and Shenker, S. Distributed Algorithmic Mechanism Design: Recent Results and Future Directions," in Proceedings of the 6th International Workshop on Discrete Algorithms and Methods for Mobile Computing and Communications, ACM Press, New York, 2002, pp. 1-13.
[26] Benjamin Szekely and Elias Torres. Ranking bookmarks and bistros: Intelligent community and folksonomy development.
[27] Segawa, O. 2006. Web annotation sharing using P2P. In Proceedings of the 15th International Conference on World Wide Web (Edinburgh, Scotland, May 23 - 26, 2006). WWW '06. ACM Press, New York, NY, 851-852. DOI=http://doi.acm.org/10.1145/1135777.1135910
[28] Aberer, K., Datta, A., Hauswirth, M.: Efficient, self-contained handling of identity in Peer-to-Peer systems. IEEE Transactions on Knowledge and Data Engineering 16(7), July 2004.
[29] Thomas Franz and Carsten Saathoff and Olaf Goerlitz and Christoph Ringelstein and Steffen Staab: SEA: A Lightweight and Extensible Semantic Exchange Architecture In Proceedings of the 2nd Workshop on Innovations in Web Infrastructure. 15th International World Wide Web Conference, 2006 (Edinburgh, Scotland)
[30] Koivunen, M. Annotea and Semantic Web Supported Collaboration, in Proc. Of the ESWC2005 Conference, Crete, May 2005. http://kmi.open.ac.uk/events/usersweb/papers/01_koivunen_final.pdf
[31] Richard Newman, Danny Ayers, Seth Russell: Tag Ontology. http://www.holygoat.co.uk/owl/redwood/0.1/tags/
[32] Smitz, P. "Inducing Ontology from Flickr Tags", paper for the Collaborative Web Tagging Workshop at WWW2006, Edinburgh, 22 May 2006
[33] Clay Shirky, 'Ontology is Overrated: Categories, Links, and Tags'. http://shirky.com/writings/ontology_overrated.html
[34] K. Aberer, P. Cudre-Mauroux, A. M. Ouksel, T. Catarci, M.-S. Hacid, A. Illarramendi, V. Kashyap, M. Mecella, E. Mena, E. J. Neuhold, O. D. Troyer, T. Risse, M. Scannapieco, F. Saltor, L. de Santis, S. Spaccapietra, S. Staab, and R. Studer. Emergent Semantics Principles and Issues. In Procceedings of the 9th International Conference on Database Systems for Advanced Applications (DASFAA 2004), March 2004.
[35] Elke Michlmayr, “A Case Study on Emergent Semantics in Communities,” Workshop on Semantic Network Analysis, International Semantic Web Conference (ISWC2005), Galway, Ireland, November 2005