Understanding documents
Reading the structure of a PDF (parsing)
What parsing produces, why a person checks it before anything else uses it, and what the warnings mean.
Structure, not meaning
Parsing turns a PDF into its parts: sections with headings, tables with cells, and figures with captions, each with its page number. It does not interpret values. Every later stage depends on this being right, so a person checks it first.
One result for each paper is current at a time. Parsing again creates a new version and keeps the previous one readable.
What a result tells you
Each result shows whether the document was a scan, and how many sections, tables and figures were found. These counts are the quickest check: open the PDF and compare.
- No sections on a full article usually means a scan or an unusual layout.
- Far fewer tables than the paper contains means the tables are images, not text.
- Warnings are listed one by one. Read them.
Approval comes first
Term recognition and extraction use only a result that a person has approved. This means no value can come from a document nobody looked at.
Approving a result does not approve the science. It says the structure is a fair copy of the document: the tables are the tables and the pages are the pages.
When the result is wrong
Reject it and parse again instead of working around the problem later. A missing table found during extraction costs a reviewer far more time than parsing again, and a fix made later is lost the next time the paper is processed.
Related
Recognising terms in a paper
How the names of proteins, properties and treatments are found in a paper and matched to the project vocabulary.
Extracting values
How candidate values are read from approved documents, what each one carries, and why a candidate is not yet data.
Uploading PDFs yourself
How an uploaded file is matched to a paper, when the match is automatic, and what happens with duplicate files.