Portfolio is an application developed by the university of Minnesota's Office of Information Technology. They recently invited me to come in and give a librarian's perspective on some of the issues and ideas that had come up regarding its taxonomy of information. These issues are of great weight as the application is rolled out to other institutions as an open-source product.
Individual "elements" in the Portfolio world are mapped to standards: the IMS ePortfolio XML spec. One of the major issues the Portfolio team was having was dividing these elements into categories and subcategories.
Categories are generally pretty clear: Your contact information will go under the category of "Personal Information," your work history will be under the category "Career", etc. But even here there are some grey areas: the element "Mentors" is under the category "Personal Information," while advisors are listed under "Education" and Informational Interview-ees under "Career"! Clearly, there is some conceptual overlap.
It gets even messier when looking at subcategories. How to fit elements under subcategories, let alone categories, that will satisfy the broad user base, extending from graduate schools to primary (Kindergarten - 6th grade) institutions?
I argued for adding a modifier to each element to allow it to be assigned multiple categories, with one set as default. These values could be edited by a system administrator but be invisible to the normal user. This relieved considerable pressure to define a universal hierarchy of element classification, a hierarchy that would need to be so universal in application as to make itself useless. The hierarchy will be used primarily for discovery - that is, for display but not for actually structuring or storing content.
Another issue was user pressure to bring in "tagging" of content. This concept is hot right now, with services like Flickr allowing users to enter free-text words describing items. We discussed using this only in a narrow sense, for users to find files that they had uploaded by using keywording. This approach of making it private, while eliminating some of the emergent properties of serendipity and conversation that the Flickr model allows, is well-suited for such a private application, and does not force the system to rely on user behavior for ensuring meaningful display to viewers.