Named Entity Recognition (NER)

Named Entity Recognition (NER) is a natural language processing technique that scans text and identifies spans referring to specific entities, people, organizations, locations, dates, or monetary amounts, and labels each with its type.

Modern NER systems are typically trained sequence-labeling models or fine-tuned language models that classify each word or token against a predefined set of entity categories. This differs from topic modeling, which finds broad, unlabeled themes across a document set: NER instead pulls out specific, named, structured facts within individual pieces of text.

NER is used to extract structured data from unstructured sources, pulling company and person names out of news articles feeding a knowledge graph, or tagging brands mentioned in social listening data, and it often runs alongside sentiment analysis so that sentiment can be attributed to a specific entity rather than the document as a whole. Common pitfalls include entity ambiguity, a model failing to distinguish a company name from an identical place name, and inconsistent performance outside the language or domain the model was trained on.