NamedEntity

A named entity refers to a real-world object that can be named, such as a person, organization, location, or temporal expression. In natural language processing, the task of identifying and classifying named entities within text is known as named entity recognition (NER). NER systems aim to locate spans of text that refer to entities and assign them to predefined categories.

Common categories include PERSON for people, ORGANIZATION for companies and institutions, LOCATION for cities and geographic features, and DATE or TIME for temporal expressions. Additional categories may cover MONEY, PERCENT, FACILITY, EVENT, WORK_OF_ART, LAW, LANGUAGE, and others, depending on the domain and annotation scheme.

Approaches to NER range from rule-based methods, which rely on handcrafted patterns and gazetteers, to statistical and machine learning techniques. Modern systems often use supervised learning with labeled corpora, employing sequence labeling models such as conditional random fields or neural networks like BiLSTM-CRF and transformer-based architectures (for example, models built on BERT). Training requires annotated text to learn how entities are formatted and contextually distinguished.

Evaluation typically uses precision, recall, and F1 score to measure how accurately entities are detected and classified. Widely used datasets include CoNLL-2003, OntoNotes, and domain-specific corpora; newer benchmarks address noisy text from social media and multi-language settings.

Applications of NER include information extraction, knowledge base population, question answering, and improved search and summarization. A related task, entity linking or disambiguation, connects recognized entities to entries in a knowledge base such as Wikidata or DBpedia, enhancing cross-document reasoning and search. Challenges include ambiguity, nested or overlapping entities, multilingual and cross-domain adaptation, and low-resource languages.