As the rest of us learned during the first tagging boom, the librarian is the natural apex predator of tagging.
As the rest of us learned during the first tagging boom, the librarian is the natural apex predator of tagging.
Turns out LC subject headings -- for all their failures -- are pretty good.
Are you using English? English words can almost mean whatever you want them to. Perhaps design your own language that removes ambiguity. Probably requires a knowledge of philosophy to distinguish between say concrete and abstract, good luck.
Maybe start with correcting the ontology of: https://cuberule.com/ (which takes a geometric approach to defining food types).
Also perhaps decide whether you want to work top-down like a directory tree (or Dewey Decimal?): resulting in standard book classification issues. Or bottom up: resulting in conflicts and discrepancies - https://news.ycombinator.com/item?id=33254025
That's what a controlled vocabulary is. It's essentially a set of tags which are clearly defined. So instead of #horses being defined purely by the word "horses," it has an attached definition along the lines of, "The category 'horses' includes equine biology, sports relating to horses, the cultural history of horses, and all other topics involving real horses. Metaphorical horses such as saw horses are not included." Tags like #horse would be redirected to #horses, since there is only one canonical horse tag in the vocabulary.
The Classifications are substantively hierarchical, though that's really an outgrowth of the fact that they're used to locate books within physical shelf space, in which a record must occupy an address (physical space), and given that the Library's settled on subject classification as its storage and retrieval basis, this maps what's effectively a folded linear structure (shelf space) onto the multidimensional subject classification. It's not ideal, but it's workable. And many of the quirks of the LoCCS come out of the fact that it addresses both the composition (comprehensive, but still US-centred) and process (shelving, search, and retrieval) of the Library.
The Subject Headings are not hierarchical, though they're structured. In particular, they're relational, with numerous subject headings referring to others. There's some parent-child relations (though the top level hierarchy is broad), numerous retired classifications, and many "use that instead of this" notes.
(I've made ... some progress ... at a structured parsing of the subject headings, though that work's been stranded Because Reasons.)