Challenges with Tagging
A lot of the success of tags and tagging can be attributed to the freedom of being able to use any word as a tag. Tagging is typically flat: all tags are at the same level and any number (or a high number) of tags can be applied to a resource. This has some drawbacks for utilising them even for the users themselves and more so for applications that aim at utilising this information automatically.
Well-known and frequent challenges with tags are that people use different words to describe the same thing, or a word has several different meanings (polysemy). People may also describe things at various levels of detail – an expert in a subject will use more detailed and specific words, whereas others use more general words. Also different forms of the same word (singulars, plurals, typos) exist. (Golder, 2006)
In addition to differences in vocabularies there are also differences between people in how they tag and why they tag. Also, applications have different restrictions and support to tags, which naturally affects the user behaviour.
There are several research papers (Golder, 2006; Maala, 2007; Marlow, 2006; Xu 2006) that report studies of the type of tags people use. In these papers the work has been based mostly on Delicious2 or Flickr3 tags. In Delicious, the following tag categories have been identified: topic, type of referenced resource, proper name (person, company, product, event, and location), subjective tags (adjectives, ratings), self reference, toDo tags and time (Golder, 2006). In Flickr photo tags categories include place, time, event, name, action and camera (Maala, 2007). The results of these studies were used when defining the tag categories for our prototype.
When the aim is to utilise tags, different types of tags give different opportunities. Topics (like travel, semanticweb, cat, cars) can be used for analyzing users’ interests as well as characteristics of the tagged resource. Proper names can be used as an indication of interests as well, particularly when additional information related to them can be found on the web. In our approach, we developed methods for automatic analysis of tag categories and methods for adding semantics to different type of tags. The aim is to use this additional metadata for finding and combining similar resources.