What is Named entity recognition?

Named entity recognition, or NER, is marking the names of things in text: people, organisations, places, dates, amounts, products and similar items, each with its type. It is one of the oldest language tasks in AI and is still used to train models that extract information from documents.

What the work looks like

You read a sentence or document and highlight each entity, choosing its type from a list. Guidelines settle the ambiguous cases: whether "Apple" is a company or a fruit in context, whether a job title is part of a person's name, how to tag nested entities such as a university named after a person, and whether to include "the" in an organisation's name.

Specialist NER

Medical, legal and financial projects use their own entity types, such as drug names, dosages, case citations or ticker symbols. These need domain knowledge and pay accordingly.

Example

"Dr Maria Lopez joined Pfizer in Madrid on 3 March 2024." The annotator tags Maria Lopez as a person (without "Dr", per the guidelines), Pfizer as an organisation, Madrid as a location and 3 March 2024 as a date. Similar tasks are listed under data annotation.