Three cheers for speculative brainstorming! Some points:
1. I can't imagine users ever wanting to tag subjects directly.
2. Although users may not tag subjects, they will tag books and books
have subjects. Because of this, LibraryThing computes relationships
between subjects and tags. So, for example, the LibraryThing tag
"globalization" relates to the following subjects, in descending order
of work-level overlap:
Globalization > Economic aspects (149)
Globalization (107)
Globalization > Social aspects (105)
Information society (99)
Diffusion of innovations (97)
International economic relations (96), etc.
The above is just "counting." There are better strategies for
evaluating the relationships between linked entities.
3. Users might be induced to confirm or dismiss these relationships,
if it were made stone-cold-easy for them.
4. LCSH has a whole "see" and "see also" system, not fully exploited
by LibraryThing and others.
5. FAST provides another way of approaching these issues.
6. I'm not sure you could incentivize users to weight correlations
between subjects and subjects, tags and subjects, etc. But I do see
benefits from baking in" some idea of relevance, even if the data is
supplied algorithmically.
7. See Fiction Finder for FAST-based clouds. I think you have to
"break apart" LCSH to produce a cloud anyone would want to see.
8. I've been thinking about foreign-language subjects. LibraryThing
has a library-data-based book-to-book recommendation algorithm, using
DDC, LCC and the subject classifications we have. These are mostly
LCSH, but not entirely. Anyway, it spots correlations no matter what
system they're in and without any attempt to "understand" them. So,
I've noticed it basing English-language recommendations on the Finnish
and Swedish headings applied to non-English editions of English works.
I have no idea how these systems work, but they encode relationships
that an algorithm can pick up on and use to improve LCSH-based
findability.
Tim
> I also wonder if user supplied _see_ and _see also_ cross references
> would be a good idea. Now I get a glimmer of how this might be useful.
> Perhaps this is really what was really being suggested in the tagging
> subject
> headings idea.
>
> LibraryThing has volunteers doing "authority" work by "linking"
> variant spellings of a single author name. While looking at this,
> it struck me that perhaps volunteers could link subject headings in a
> similar way.
> Saying that LCSH 1 is very similar to LCSH 2. Perhaps
> there could be a way to let people assign a "similarity number" to the link,
> with 5 being very similar and 1 not so much. As you know from browsing
> catalogues, similar or related subject headings show up because of links to
> a particular title.
> Or because your search keywords retrieved titles that have similar LCSH's.
> This would be a way to link SH's without the title(s). Could you then make
> LCSH "clouds?"
>
> Anyway, these are all just musings, but I wonder what you think of these
> ideas?
> Brainstorming, you know...
>
> Kevin
>
Received on Fri Mar 02 2007 - 05:46:53 EST