✦ScienexySCIENCE KNOWLEDGE LIBRARY
Science reference · SCI-CS-003-129

Named Entity Recognition

Named Entity Recognition (NER) is an NLP task that identifies and classifies named entities such as people, organizations, locations, dates, and other meaningful expressions in text.

Precision = TP / (TP + FP)
Recall = TP / (TP + FN)
F1 = 2 × (Precision × Recall) / (Precision + Recall)

Explanation

Named Entity Recognition (NER) is an information-extraction task in natural language processing. Its purpose is to locate meaningful entity expressions in text and assign them to predefined categories.

Common entity categories include:

1. PERSON: Names of people.
2. ORGANIZATION: Companies, institutions, and other organizations.
3. LOCATION: Countries, cities, regions, and geographical locations.
4. DATE: Calendar dates and periods.
5. TIME: Times of day or time expressions.
6. MONEY: Monetary amounts.
7. PERCENT: Percentage expressions.

For example, in the sentence 'Sundar Pichai visited Google headquarters in California in 2024', an NER system may identify 'Sundar Pichai' as a person, 'Google' as an organization, 'California' as a location, and '2024' as a date.

NER systems commonly use sequence labeling methods. Traditional approaches include rule-based systems, dictionaries, and statistical models such as Conditional Random Fields. Modern systems often use neural networks and Transformer-based contextual representations.

A common tagging scheme is BIO:

- B-: Beginning of an entity.
- I-: Inside an entity.
- O: Outside an entity.

For example, the words in 'New York City' may be tagged B-LOC, I-LOC, and I-LOC. This allows a system to represent multi-word entity spans.

NER is challenging because the same word can represent different entity types depending on context. New entity names, spelling variations, ambiguous abbreviations, and domain-specific terminology can also reduce accuracy. NER performance is often evaluated using precision, recall, and F1-score, with exact entity-span matching commonly used.

Variables

SymbolMeaning

TP

Correctly identified entities or entity predictions under the evaluation scheme

FP

Incorrectly predicted entities

FN

Entities that the system failed to identify

Precision

Proportion of predicted entities that are correct

Recall

Proportion of reference entities that are identified

F1

Harmonic mean of precision and recall

Example or application

An NER system processing 'Microsoft opened an office in Dhaka' can identify 'Microsoft' as an organization and 'Dhaka' as a location.

Text

NER detects entity spans in text and assigns labels to them. Common categories include person, organization, location, date, time, money, and percentage.

Key Points

  • NER identifies entity spans in text.

  • Entities are assigned predefined categories.

  • Common categories include people, organizations, locations, and dates.

  • BIO tagging represents the boundaries of multi-word entities.

  • Context helps resolve ambiguous entity names.

  • Precision, recall, and F1-score are common evaluation measures.

Applications

  • Information extraction

  • News analysis

  • Search engines

  • Document indexing

  • Customer-support analytics

  • Financial document processing

  • Medical text analysis

Learning Outcomes

  • Define Named Entity Recognition.

  • Identify common entity categories.

  • Explain the BIO tagging scheme.

  • Understand how NER systems are evaluated.

Practice Questions

  • What is Named Entity Recognition?

  • What is the difference between PERSON and ORGANIZATION entities?

  • How does BIO tagging represent a multi-word entity?

  • Why is context important for NER?

Related Concepts

  • Tokenization

  • Part-of-Speech Tagging

  • Sequence Labeling

  • Information Extraction

  • Transformer Models

Common Mistakes

  • Assuming every capitalized word is a named entity.

  • Confusing entity classification with document classification.

  • Ignoring entity boundaries when evaluating NER.

  • Assuming every NER dataset uses exactly the same entity categories.

Sources and verification

Sources are shown as provided in the project data. Not every source has yet been independently verified. Please report errors or missing citations.

  • Nadeau, D., and Sekine, S., A Survey of Named Entity Recognition and Classification, Lingvisticae Investigationes, 2007.
  • Jurafsky, D., and Martin, J. H., Speech and Language Processing, 3rd edition draft.

Suggest a correction or source · More in Artificial Intelligence & Machine Learning