Picking out the names of things in text, such as people, organisations, places, dates and amounts, and labelling each with its type. It turns free text into structured data: from a news story, it can list who was mentioned, where and when. It is often the first step in building a searchable database out of documents.
Technical view
Named Entity Recognition (‘NER’) is usually framed as sequence labelling, where each token is tagged as the Beginning, Inside or Outside (‘BIO’) of an entity span. Approaches moved from hand-written rules and name lists, through conditional random fields, to fine-tuned transformer encoders, and large language models can also extract entities from a prompt with no task-specific training. Hard cases include nested and ambiguous entities, unseen names, and domain types such as drugs or part numbers. Recognition is usually followed by entity linking, which resolves each mention to a record in a knowledge base.
Related
Part of AI Terminology Index, section: Modalities and Applications.