Home »

Semantic Tagging

Wednesday, September 19, 2007
Tagging in Web2.0 is the ability to associate a keyword like “food”, or set of concatenated keywords like “mycooking” with a resource such as a blog entry or a photograph. The benefit of doing this is that the interface then can allow browsing by tag, and can show the popularity of certain tags in a “tag cloud.”6 A tag cloud is a list of all the tags used in the current context, such as one person’s blog, where the size of the individual tag’s font size is proportional to its usage. For example, a blogger may frequently write about food and restaurants, so every blog entry about food is tagged with “food.” The blogger, however, may want to specify when the food described is their own creation, so they use the tag “myfood”. They can then quickly tell, by looking at the size of the tags in their tag cloud the proportion of posts about “food” compared to the posts about “myfood”. While this knowledge may be useful, the ability for a computer to automatically utilize the tag “myfood” with other data is limited. However, if one were to describe the meaning of this tag, by referencing ontologies, the meaning, to both computer and humans alike becomes clearer. This not only means that ambiguity of terms is reduced, but that the users are in control of their own vocabulary.

We propose that a Semantic Tag is one where the tag itself is backed up by an RDF graph, functioning therefore as a useful shortcut to link knowledge to content. This is better than a keyword search as the Semantics are backed up with an Ontology, allowing the knowledge to not only be integrated easily with other Semantic data, but also the inference over this Ontology. For example, someone may tag something with “Beef Steak”, and this infers, from the relevant food ontology, that another person is describing with the tags Beef, Meat, and Food. This means that someone who then searches on non-vegetarian cuisine will find this item. Semantic Tags should also be subsumptive, for example, the tag "mSpace" also infers the tags "Semantic Web Research" and "HCI Research" and the tag "Semantic Web Research" infers the tag "Computer Science Research". The RDF behind these tags depends on the usage, and therefore needs to be alterable by users. Likewise, users should be allowed to alter the meaning of a tag, by adding, removing and altering triples before they publish their use of the tag.

Beckett describes a practical way for tags to be formatted for semantics in current tag applications. This is a way to add semantics to tags without alteration of current applications, by specifically namespacing tags. For example, if a user was to specify that a file's type is MP3, they would tag it with "system:filetype:mp3". The problem of how to describe the underlying concepts and how to resolve these is left open. Alternately, our approach requires a small alteration to tagging interfaces, while also providing a method to resolve the knowledge where it has been previously described, and a way to allow the bootstrapping of knowledge in the event that it has yet to be properly described in an ontology, after the tagging has occurred. The parameters of these edits presents an interesting problem for shared meaning of tags since spam, vandalism and point-of-view differences give people reason to alter the meaning of a tag in a way that is not shared by all people who use that tag. We suggest a system where all possible permutations of the knowledge of a tag are all stored, and whenever a user specifies the tag, they are given the choice of which meaning to use, ordered potentially by how many times that meaning has been used. For example, the use of the tag TonyBlair could have many people tagging with the meaning of “Prime Minister” and “Politics”, but there may also be a number of users that also wish this tag to infer “Liar”. This option should not be suppressed7, however it may be more convenient for the most utilized tag meanings to be at the top of a selection list, while of course enabling the preferred and perhaps less popular selected meanings of a tag to be immediately apparent when reading the meanings as they appear in a blog post which uses them.8 The tag "TonyBlair" may well likely be vandalized (associating the tag with something insulting), however not all tag users would share this assessment, or want to include it in their use of the tag even if they did, so the alteration of the tag's meaning by lightweight interaction, such as deselecting this attribute of a tag can act as a vote against that meaning, and would help to reduce the occurrence of that meaning being associated with that tag. Wikipedia has faced a similar problem – It allows editing of its articles by anybody, and as such has to deal with problematic entries, Spam and difference of opinion on a regular basis. In spite of this, it reigns as a resounding success, based mainly on its policies for dispute resolution (38) that offers the ability for users to vote on whether article should be deleted, altered, or how a particular topic is described. With enough votes against a piece of knowledge behind the semantics of the tag, that vandalism will be removed.
Tags: , , , ,

Comments[ 0 ]


Post a Comment