Open pnb opened 2 years ago
I thought for a while that this might be specific to DOCX files, but apparently LaTeX papers can have manually added \bibitem
references, including mismatched references (e.g., ps9itMnZ4n), so it should be handled for LaTeX as well.
Might try custom training anystyle on the EDM style in particular.
LFfQRMVQeF is one example with [0, 2]
Fixed the in-text cite parts in e484154, which handles basically all the problems with false positives related to numbers in math ranges and other oddities. There are certainly more cases that will crop up, but this works for all accepted papers from last year now at least.
Still does not handle false positives in the anystyle part though.
One more type of in-text false positive in xSIDvwju3D, with [6, 25–27]
type of formatting.
Now partially implemented, but only for JEDM papers.
Per #6, there are some tricky cases with reference checking, including ranges in equations and others. The anystyle package used for reference checking may merit improvement as well, since it has some false positives and other misses like confusing journal volumes for page numbers (in which case one thing is still missing, but the wrong thing is identified).
This might require some custom parsing and context checking -- e.g., whether the current brackets are in a MathJax tag and should be ignored.