Repository navigation
Under 0.5.0 a quote has to start and end on whole words. That exposed text the extractors had welded together, where a correct quote now failed. This release stops the welds (#163).
Fixes
- Crossref abstracts (DOI). JATS titles, sections and paragraphs, plus HTML
divandbr, are now blank-line-separated paragraphs. Before,<jats:title>Abstract</jats:title><jats:sec><jats:title>Disease overview</jats:title><jats:p>Acute…becameAbstractDisease overviewAcute. - PubMed abstracts. A section heading welded onto its sentence becomes a
Heading:label. That covers inline<b>Background</b>Neuroblastomaand headings PubMed stores with no markup (exposures.ConclusionsPoison). Only listed, title-case headings directly followed by a capital letter count. Bold that isn't welded, such as<b>IL</b>-6, is left alone. - PMC XML citation markers. A
<xref ref-type="bibr">, or a<sup>holding only such markers, now gets a space before it. SoFOXL2<sup>8</sup>no longer reads asFOXL28, andPGE2.<sup>41</sup>no longer as the decimalpge2.41. It also gets a space after it, except before a dash, so a range like[1-3]keeps its dash. - PMC XML lead-in labels. A paragraph's bold label welded to the next word (
<bold>R 2.6</bold>We recommend) is set off from it. - Matching. The trailing-citation allowance also accepts superscript digits, so
small earsmatchessmall ears¹⁻³. As before, it applies only after a quote ending in a letter.
Existing caches are not refreshed automatically
EXTRACTOR_CACHE_VERSION is unchanged on purpose. Bumping it would re-fetch every cached reference on the next validation. So an entry cached before 0.5.1 keeps its welded text until you re-fetch it with linkml-reference-validator cache reference <ID> --force, which means results depend on when an entry was cached.
Check what a forced re-fetch gives back before keeping it. Full text that came from a PDF or publisher HTML can come back as an abstract only. In one re-fetch of 453 references, 66 came back abstract-only or empty.
Full changelog: v0.5.0...v0.5.1