Recognising terms in a paper

What it does

Term recognition (also called entity recognition) reads an approved document and marks where it names a protein, a functional property, a treatment or a condition. Each mention keeps the text around it, so you can judge it without opening the paper.

Mentions are then matched to the project vocabulary. A mention that matches clearly helps extraction later; one that does not is shown as uncertain.

It creates no data

This step creates no values and no dataset rows. It prepares the names of proteins, properties and treatments that extraction relies on. A missed mention makes extraction harder, but it cannot by itself put a wrong number into the dataset.

Reading the list

Each paper shows how many mentions were found, matched and left uncertain. The latest attempt is shown even if it failed, so a failed attempt is not hidden behind an older result.

• Many mentions but few matched usually means the vocabulary is missing a synonym.
• No mentions on a parsed article usually means no body text was read. Check parsing first.
• Uncertain mentions are not errors. They feed the vocabulary review.

Improving the matches

The fix for poor matching is almost always the vocabulary. Add the synonym the paper actually used, then run term recognition again. That helps every later paper in the project, and a colleague can review the change.