Tags: What are They Good For?Prentiss [email protected]://prentissriddle.comINF 385Q: Knowledge ManagementSchool of InformationThe University of Texas at Austin5/12/2005Tags: What are They Good For?1IntroductionTags and folksonomies represent a novel approach to the problem of organizinginformation. These terms refer to the use of uncontrolled user-supplied textual labels ("tags") tocategorize and identify resources in a nonhierarchical shared information space (a"folksonomy").Although the concepts of uncontrolled keywords and user-supplied metadata are not new,the particular environments in which they are being used have generated intense interest; as ofthis writing Google shows 273,000 entries for the term "folksonomy," a neologism which did notexist a year ago. Yet at this time there are just two relatively mature and heavily used systemsbased on these ideas, the bookmarking system del.icio.us (http://del.icio.us) and the photosharing system Flickr (http://flickr.com). That these two systems are the exemplars on which theenthusiasm for folksonomy rests is intriguing, because del.icio.us and Flickr are different enoughto show that not all folksonomies work in the same way and yet they are similar enough not tonecessarily bound the space in which tagging might be useful.This paper will briefly state some of the strengths and weaknesses of tagging that havebeen demonstrated by existing systems, but its focus will be to speculate about the larger, as yetunanswered questions regarding other domains in which tagging may or may not prove useful.What we know so farMillions of words into the folksonomy frenzy, what do we know? Existing systems havedemonstrated that tagging is good for a handful of things.Tagging is useful for personal recall, or finding again what you have seen before. Thisis the main purpose of the bookmarking system del.icio.us and its many less widely usedimitators (see Irox (2004) for a comparison of several).Tags: What are They Good For?2Tagging supports social effects. One frequently noted example is the common practicein Flickr of agreeing on a tag and convening an informal group of users who create new photosexpressly to use it. The canonical instance of this practice is the tag "squared circle"(http://flickr.com/photos/tags/squaredcircle/).Tagging promotes serendipity, or the pleasant and sometimes useful discovery of theunexpected. Prowlers of libraries and used bookstores know that the collocation of books on ashelf by topic creates interesting juxtapositions of items which are sufficiently different that theinformation seeker would not have thought to explicitly search for them. Tagging heightens thiseffect by allowing items to be "collocated" on any number of serious or whimsical criteria,including the identities of the users who tagged them, and by increasing the factor of chancethrough the imprecision of tags.Tagging is good for novelty, or what del.icio.us creator Joshua Schachter calls"interestingness" and defines as "the first derivative of popularity" (Weinberger, 2005, p. 20).Schachter's statement implies that ideas or resources which are already widely known becomeuninteresting through familiarity and that items which are gaining in attention at a given time arethose most likely to provoke our interest. Tags provide a convenient way to segment andmeasure the rise and fall in popularity both of individual resources (how many times they havebeen tagged) and the categories they fall into (the tags themselves). This focus on the new ispresumably why the default presentation of items in del.icio.us and Flickr is in reversechronological order and it dovetails nicely with the common support of RSS feeds in taggingsystems.We also know that tagging (at least as it stands now) is not good for several other sorts ofinformation discovery and retrieval.Tags: What are They Good For?3Tagging does not perform well at measures of precision and recall (in a different sensefrom "personal recall" above). In information retrieval systems, precision is a measure of theelimination of false positives, that is, the portion of retrieved documents which are relevant.Recall is a measure of the elimination of false negatives, that is, the portion of relevantdocuments in the system which are retrieved (Korfhage, 1997, p.194). Because users can applymany different tags to a single concept or a single tag to many different concepts, the results fora given query are likely to be both noisy (low in precision) and incomplete (low in recall).Tagging is not good for ontologies in the sense of precisely defined relationships amongconcepts. The most familiar sort of ontology in the domain of information discovery andretrieval is a hierarchical system for classification by subject with a controlled vocabulary and athesaurus defining broader and narrower terms. The uncontrolled vocabulary of a taggingsystem is essentially the polar opposite of a system like the Library of Congress SubjectHeadings or the Dewey Decimal System (Mathes, 2004). Similarly, free tags do not provide thestructure necessary to capture and enforce compliance with other kinds of ontologies, fromfaceted classification systems to the schemas being constructed to support the Semantic Web (forexample those at http://www.schemaweb.info/). It is true that users of del.icio.us or Flickr cancreate their own conventions to represent hierarchies or other relationships, sometimes expressedby internally segmenting a tag with punctuation, but since there is no formal mechanism forrecording or validating those conventions and since the tags are recorded in a common spacewith overwhelmingly noncompliant tags, such effects are local.What we don't know yetThere is a longer list of things about tags and folksonomies which we do not yet know. Iwill discuss these questions of their possible utility in terms of the domains to which they may beTags: What are They Good For?4applicable, the social and organizational contexts in which they may work, and the degree ofstructure which may be imposed on them.DomainsBesides bookmarks and photos, what other domains might tagging be good for?People. If folksonomies are this year's great fad, it could be said that last year's wassocial network systems. Opinions vary as to the success of commercial social network servicessuch as Friendster and Orkut, but an area of continued interest is distributed social networksystems expressed in XHTML (XFN, n.d.) or XML schemas such