The papers mentioned below may have information related to your question.
Hi dear Andrew,
I am very interested in this issue as well.
If I am not wrong, it relates to some studies by the Topic Maps community on extracting metadata from unstructured data or unstructured sources applying automated classification (groupping different 'keywords' with their "subject" which would have the identifier). I found this presentation very useful: http://www.topicmapslab.de/publications/topic_maps_and_automated_classification
Also, there are some studies in other communities that may have the perspective you are describing. They are usually referred to as "emergent semantics", and usually use facets to group conceptually those taggs from a bottom-up analysis.
This is a very interesting thesis that gives ideas and an overview of these studies: https://oda.hio.no/jspui/handle/10642/313
And this one (in Portuguesse) also tries to relate groups of tags with their correspondent Dublin Core property (and found that some don't map at all, but proposed another metadata application profile that could include them):
http://hdl.handle.net/1822/9564
Also these articles may be helpful:
http://www.biomedcentral.com/1471-2105/10/313
http://onlinelibrary.wiley.com/doi/10.1002/asi.21190/abstract
It would be interesting to know if you find more on this applied to Topic Maps.
Regards,
Liliana Melgar
2010/12/4 Andrew S. Townley <[log in to unmask]>
>
> Hi Folks,
>
> Been doing some searching, but haven't found much detail about practical applications of using existing folksonomy classifications (e.g. tagging) to derive or infer subject identity in varying degrees of relevance. Most of what I saw on "semantic tagging" or anything similar I could think of a search kewords has the opposite: provide a way to more precisely tag resources using the proper semantics, vs. the bottom-up approach that I'm looking for.
>
> Has anyone seen anything about this related to Topic Maps or have people trying to map large bodies of existing information done anything in this space already?
>
> I have a few thoughts about how something like this could work to generate degrees of certainty in subject identity, but they're just thoughts
>
> Anyone doing this sort of thing with Topic Maps?
>
> Cheers,
>
> ast
> --
> Andrew S. Townley <[log in to unmask]>
> http://atownley.org
> _______________________________________________
> topicmapmail mailing list
> [log in to unmask]
> http://www.infoloom.com/mailman/listinfo/topicmapmail
--
Liliana