Because I am a complete another geek, I'm fascinated by an alternate database approach called Neo4j, which focuses on modeling data as graphs as a first order operation rather than as an emergent phenomenon. They recently had an article linked from their newsletter which actively discussed developing a hierarchical taxonomy from a set of unstructured folksonomic tags, which I find really interesting from a mechanical point of view.
That is to say, building a hierarchy which is built from the ground up from the tags people actually use.
Something alongside a way to tell the system that you are less interested in a given topic/subject/tag, this would mean that usage would dictate the taxonomy, to some degree and the user could have a tool to focus their attention on just the part of the content they're interested in.
Because, after all, "photography" as a tag still applies to all of those pictures of every camera review and picture of their child. It is all photography. There's no question about that. And for macro photographers, their work is still photography even if it is also likewise the more specific macro photography.
This provides more of an argument for a system which allows users to provide a certain amount of weight to any given topic that they follow. Apocryphal may be interested in photography in general, but is really interested in macro photography. If attached to their account, their list of interests was something like "Photography+, Macro_Photography++" and the system knows to order things with more pluses more highly, then when looking at their stream of posts which are interesting, the ones that are most likely to be interesting are closer to the top.
This is strangely a pretty good argument for removing the limitation of only five topics on a given post because eventually the system will have more than five topics deep that certain parts of the conversational taxonomy might need to be a part of.
(Though that might be minimized if the presence of a tag implies some degree of the expressed interest by the user transfers up and down the hierarchy to some degree. If the next step up the hierarchy for a topic that a user expresses interest in gets considered effectively 25% of its sub-topic, and topics downstream of the selected interest topic receive 50% (unless otherwise specified by the user at some point), you could get a very effective automatic sort from a very limited number of expressed interests.
Just spit-balling, here.