presentsUNAIVERSE
The UN AI timeline

    Learn AI /Computer Vision and Language Technology /Named entity recognition

    Named entity recognitionNER

    Also called: NER, entity extraction.

    Detail

    In five seconds

    What it is

    A text-processing method that finds mentions such as people, organisations, places, dates or amounts and labels each mention by type.

    In plain words

    It is a way for software to scan text, mark pieces like names, places and dates, and turn them into labels that other tools can use.

    An everyday example

    Imagine highlighting a message with one colour for people, another for places, another for dates and another for prices. The words stay where they are, but each relevant span receives a category. Context matters: “May” might be a person’s name in one sentence and a month in another. Named entity recognition performs a similar tagging task automatically.

    Where the story stops working

    A student usually applies clear category rules consciously and may use world knowledge from outside the page. An NER system does not reliably do that. It detects statistical patterns or follows programmed rules, and its label set may be narrower, inconsistent across languages, or wrong when context is limited.

    Remember thisNamed entity recognition finds spans of text and assigns predefined types such as person, organisation, location, date or amount. Context helps determine the label, but the output still requires checking.

    Understand it

    Understand it

    Why it matters

    Many important documents are mostly free text. If software can reliably pull out names, places, dates and amounts, teams can search faster, summarise trends, route work and spot missing information. For non-specialists, the main value is practical: NER can reduce manual sorting and support downstream analytics. The main decision questions are whether the labels match your task, whether performance is good enough in your languages and document formats, and what human checking is still needed. A system that works well on English news may perform poorly on local place names, multilingual reports or scanned text with OCR errors.

    How named entity recognition turns text into labelled spans

    After viewing the diagram, the reader should be able to say that NER takes raw text, finds exact word spans, assigns entity types and returns labelled text for later use.

    1. 01

      Input text

      Start with an ordinary sentence or document.

      Show one sentence in a rounded rectangle: “Amina Hassan met UNICEF staff in Nairobi on 12 June 2025.”

    2. 02

      Split into tokens

      The system first breaks the text into manageable units, usually words and punctuation, often called tokens.

      Show the same sentence with each word in a separate pill-shaped box, connected lightly to the input.

    3. 03

      Find entity spans

      The system groups one or more tokens that appear to form a named entity mention.

      Draw coloured brackets around “Amina Hassan”, “UNICEF”, “Nairobi” and “12 June 2025” across the token boxes.

    4. 04

      Assign labels

      Each detected span receives a type such as person, organisation, location or date.

      Place small coloured tags above each bracket: PERSON, ORG, LOCATION, DATE. On the far right, show a simple table with two columns: text span and label.

    A left-to-right flow with one sample sentence moving through four stages: raw text, tokenised words, detected spans, and final labels. Read from left to right to see how the same sentence is gradually annotated rather than rewritten.

    Hovering or tapping a coloured span highlights the matching label in the output table and vice versa.

    Describe this without the diagram

    The diagram shows a sentence moving through four stages from left to right. First is the plain sentence: “Amina Hassan met UNICEF staff in Nairobi on 12 June 2025.” Second, the sentence is split into separate word boxes. Third, brackets group four spans: Amina Hassan, UNICEF, Nairobi, and 12 June 2025. Fourth, each span has a label: person, organisation, location, and date. The main point is that named entity recognition does not rewrite the sentence. It identifies exact stretches of text and attaches category labels to them.

    With reduced motion: A static four-column layout shows the same sentence at each stage: plain text, token boxes, grouped spans and labelled output table, with matching colours used instead of animation.

    Named entity recognition takes text as input, such as an email, report, article or transcript. It looks for spans of words that refer to specific kinds of things, often people, organisations, locations, dates, times, money values or product names. The output is usually the original text plus labels attached to those spans.

    For example, in the sentence “Amina Hassan met UNICEF staff in Nairobi on 12 June 2025”, a system might mark “Amina Hassan” as a person, “UNICEF” as an organisation, “Nairobi” as a location and “12 June 2025” as a date. Some systems only find and label the spans. Others also link them to a database entry, but that extra step is different and should not be assumed.

    What happens inside depends on the method. Older systems often used hand-written rules and dictionaries. Newer systems often use machine learning, meaning they learn patterns from many labelled examples. They may use clues from spelling, nearby words, sentence structure and the wider context.

    NER is useful because it turns unstructured text into fields that software can sort, count, search or pass to later steps. But it is not the same as understanding the whole text. A model may correctly label a name while still missing sarcasm, uncertainty, who did what, or whether two mentions refer to the same real-world person.

    How “Colour-coding a message” maps onto the real thing

    In the storyIn the AI system
    The practice exam essayThe input text
    Different highlighter coloursEntity labels such as person, location and date
    Marking exact words in the essayFinding the span of text that forms the entity
    Using the surrounding sentence to disambiguate “Jordan”Using context to choose the correct entity type

    A worked example

    Reviewing the entity labels produced from one sentence in a side-event concept note.

    Starting from: “The concept note mentions UNESCO, WHO and Kathmandu.”

    1. Split the sentence into tokens while retaining their positions in the text.
    2. Identify the spans “UNESCO”, “WHO” and “Geneva” as candidate entities.
    3. Use the surrounding words and learned or rule-based patterns to assign ORG to UNESCO and WHO, and LOC to Geneva.
    4. Return the spans, labels and text positions in a structured form for search or review.

    Result: One possible output is: [UNESCO: ORG], [WHO: ORG], [Geneva: LOC]. The labels describe the mentions in this sentence and may differ where the same words appear in another context.

    In a United Nations workplace

    At ITU in Kathmandu, a duty officer is finalising a side-event concept note before the front office sends it for clearance at close of business. A document tool is used to identify organisations and locations mentioned across earlier, approved concept notes.

    Named entity recognition can turn those mentions into structured fields, helping the duty officer compare proposed participants, venues and institutional references without searching each document manually. The extracted labels remain prompts for review, not authoritative entries for the cleared text.

    Before you try thisDo not upload uncleared drafts to an external tool unless it is approved for that material. Names may also be missed, split incorrectly or assigned the wrong type, so every extracted field should be checked against the source text before circulation.

    Where you may hear it

    • During a vendor demo for a document-search tool
    • At a technical review of an information-extraction pipeline
    • In a working session on tagging situation reports
    • During clearance of requirements for a knowledge management system

    Avoid confusion

    Avoid confusion

    Often confused with

    Common misconceptions

    • NER identifies every important fact in a document.

      NER usually identifies only certain categories of mentions, such as names, places or dates. Important facts like obligations, risks, opinions, relationships or causes often require other methods or human reading.

    • If a system finds a name, it knows exactly which real-world person or organisation that name refers to.

      Finding and labelling a mention is different from linking it to a specific record. “Paris” could be a city or a person’s name, and “Ministry of Health” could refer to different organisations in different countries.

    • NER is just keyword matching.

      Simple systems may use lists and rules, but many NER systems also use context. That matters because the same word can have different roles in different sentences, and entities can contain several words.

    • High accuracy on one dataset means the system is reliable everywhere.

      Performance often changes across languages, sectors, writing styles and document quality. Abbreviations, OCR mistakes, uncommon names and low-resource languages can reduce reliability.

    Quick check

    Which output best matches named entity recognition?

    Go deeper

    Go deeper

    The simple version

    It is a way for software to scan text, mark pieces like names, places and dates, and turn them into labels that other tools can use.

    The technical version

    Named entity recognition is a sequence labelling or span classification task in natural language processing. Given text, the system identifies token spans that correspond to predefined entity classes and assigns a label to each span, such as PERSON, ORG, LOC, DATE or MONEY. Implementations may be rule-based, dictionary-based or learned from annotated corpora. In learned systems, the model uses features or internal representations derived from token form, surrounding context and sometimes sentence-level structure. Output may be represented with span boundaries, token-level tags such as BIO tagging, or a structured list of entities. NER is distinct from entity linking, which maps a mention to a specific knowledge-base entry, and from relation extraction, which identifies how entities are connected.

    Reviewed external resources

    • EncyclopaediaNamed-entity recognitionWikipedia

      Wikipedia gives a broad overview of named entity recognition and suits readers who want a quick orientation and basic terminology.

      Intermediatelink checked 2026-07-31

    • ArticleMachine Learning GlossaryGoogle

      Google’s Machine Learning Glossary is useful for readers who need short definitions of related technical terms they may encounter while learning about NER.

      Intermediatelink checked 2026-07-31

    • CourseMachine Learning Crash CourseGoogle

      Google’s Machine Learning Crash Course gives general machine learning foundations that help technical beginners understand how systems like NER are trained and evaluated.

      Intermediatelink checked 2026-07-31

    • VideoVideo explanations: Named entity recognitionYouTube search

      The YouTube search results page offers multiple video explanations of named entity recognition and suits readers who prefer to compare different teaching styles before choosing one.

      search results, not one pagelink checked 2026-07-31

    • DiscussionCommunity discussion: Named entity recognitionReddit search

      The Reddit search results page shows community discussions about named entity recognition and suits readers who want informal questions, examples and practitioner experiences, with careful judgement about reliability.

      search results, not one pagelink checked 2026-07-31

    Where this came from

    Drafted with a language model through an offline pipeline, checked against this site's schema, and every external link above fetched to see whether it answers. Treat it as a starting point and check anything that matters against the primary sources. How this site is built and what that means.

    The one thing to remember

    Named entity recognition finds spans of text and assigns predefined types such as person, organisation, location, date or amount. Context helps determine the label, but the output still requires checking.