Problems in tag-based item classification
Tuesday, 13. March 2007, 04:56:48
Tags are in... but that doesn't mean that they are better in every aspect compared to the traditional hierarchial system.
Let's start by comparing tagging with the hierarchial way of organizing things
Lets say you want to organize your photos. If you go with a hierarchy, it could look something like this
photos/summer/greece - photos of the summer holiday in Greece
photos/me - photos of yourself
photos/summer/greece/me - photos of yourself in greece
or maybe photos/greece or anything like that.
This way you could be sure that all your photos from summer holidays would be found in photos/summer, but there's also a problem: You would have to have the photos in multiple places if you wanted to have all the photos of you in photos/me and all summer pics in photos/summer because there would probably be pics of you in the summer too...
Also, if you'd start looking for pictures of yourself in Greece, looking in photos/me would be pointless if you had put them in photos/summer/greece.
Now, with tags you could just give every photo all the keywords that fit it. Photos of you in Greece could be tagged with "me", "Summer" and "Greece" and so on.
This way you can more easily define "keywords" for finding the photos, too.
The problem arises when you try to find things: When you search for "me", did you really find all the photos you're looking for?
You may have to broaden your search and search for "me", "summer", "greece" etc.
Another good example is that if want to search for links in del.icio.us about cars, you might not find all intresting links with just "cars". You might have to search for "cars and ford and aston martin" and so on.
In a hierarchial system you'd just look in cars and ford and aston martin would be listed under it in addition to other items related to cars.
Other problems are synonyms and spelling. For example, both television and tv could be used as a tag for an item about television. Or if you search for "car" items tagged with "cars" won't be returned even though they are very likely related to your search.
There are some tagging systems around which let administrators to define synonyms which solves one problem quite efficiently, but that means admins would have to keep track of all the tags used in the system which may prove difficult especially with many many users.
One option would be letting users "tag tags". This could be used to create obvious links to related tags and to create a hierarchy. This could lead to problems when a malicious user decides to screw with the tags taglist.
Another option would be to go with an admin/moderator managed list of synonym and hierarchies. Also, del.icio.us for example, provides "related tags" which seems like a guess depending on what tags have been used the most times on the items found in addition to the tags searched for.
Personally I think that tags are a much better alternative to traditional hierarchial models for categorizing things. However, a hierarchial view provides a very good model of how things are related to each other.







How to use Quote function: