grobidOrg / grobid / 34920435922

43%

master: 44%

Pull #1560

github

fix: empty trailing reference discarded all citation training data

When the reference segmenter ended the section with a bare label (e.g. a
number split off the last DOI) and no reference text, getExtractionResult
still emitted a result with an empty text. CitationParser.trainingExtraction
then returned null for the whole batch, so createTraining silently skipped
training.references.tei.xml and training.references.authors.tei.xml.

With the article/footnotes-refs DeLFT reference-segmenter model this happens
on the PDF of mpilhlt/grobid-footnote-flavour#43.

- drop a trailing chunk without reference text, as already done for the
  other references
- skip blank inputs in trainingExtraction instead of aborting the batch

10193 of 26390 branches covered (38.62%)

Branch coverage included in aggregate %.

21332 of 47353 relevant lines covered (45.05%)

1.83 hits per line