Near as I can tell, the best that's available is starting with the class(es) you want (e.g., https://id.loc.gov/authorities/classification/H.html https://id.loc

Near as I can tell, the best that's available is starting with the class(es) you want (e.g., https://id.loc.gov/authorities/classification/H.html https://id.loc.gov/authorities/classification/H.html for social sciences) and crawling recursively to grab whatever serialization of the RDF you want to consume. There's no SPARQL endpoint for id.loc.gov, alas, but it beats resorting to the PDFs.