What is named entity recognition (NER)?
Erwin Berkouwer
Named Entity Recognition (NER) is an NLP technique that automatically identifies and labels names of people, organizations, locations, dates, and amounts in text, via rule-based methods or trained language models. Relevant for SMEs to automatically structure invoices, contracts, and emails, and improve search functionality. Limitations exist with unusual names, abbreviations, and industry jargon; human review remains advisable for financial or legal applications.
Named entity recognition (NER) automatically detects names, organizations, dates, and amounts in text. Useful for structuring documents and emails.
Named Entity Recognition (NER), also known as entity recognition, is an AI technique that automatically identifies and labels specific elements in text, such as names of people, organizations, locations, dates, and amounts. Rather than just reading text as a string of words, NER extracts the meaningful "entities" that matter for further processing.
The technique is part of Natural Language Processing (NLP) and often forms a crucial first step before other AI applications, such as data extraction or automation, can work.
How named entity recognition works
NER models analyze text and assign labels to words or phrases. A sentence like "John Smith from Smith Bakery in London ordered on June 3rd for 450 euros" would be recognized as:
- Person: John Smith
- Organization: Smith Bakery
- Location: London
- Date: June 3rd
- Amount: 450 euros
There are different ways NER models do this:
- Rule-based: fixed patterns and word lists recognize known names and formats (for example, date formats or postal codes).
- Machine learning / language models: the model is trained on large amounts of labeled text and learns to recognize patterns itself, even for names or organizations it hasn't seen before.
The output of NER is structured information from unstructured text: exactly what's needed to make free text usable for systems, dashboards, or automation.
Why it matters for SMEs
Many business processes still run on free text: emails, contracts, invoices, notes, customer messages. That text often contains exactly the information a system needs, but in a form that isn't directly processable.
Named entity recognition helps convert that text automatically into usable data:
- Processing invoices and documents: automatically extracting amounts, dates, and supplier names without manual entry.
- Structuring customer communication: recognizing names, product names, or locations in incoming emails or tickets.
- Improving search and filtering: making large volumes of documents searchable by specific people, companies, or time periods.
[Estimate]: for companies spending a lot of time manually retyping data from emails, PDFs, or forms, named entity recognition can eliminate a large part of that manual entry work.
A practical example
An accounting firm receives invoices in all kinds of formats and layouts from clients. Instead of manually retyping every amount, date, and supplier name into the accounting software, a NER model automatically recognizes this data from the incoming text or scanned document. Only deviating or unclear cases still get checked manually.
When it works, when it doesn't
Named entity recognition is powerful, but not always the best choice.
| Good fit | Less suitable |
|---|---|
| Structuring large volumes of documents or emails | Very small amounts of text where manual work is faster |
| Processing invoices, contracts, and forms | Extremely informal or highly deviating writing styles |
| Improving search in archives | Situations requiring 100% accuracy immediately without checks |
| Preparation for further automation | Languages or domains the model hasn't been trained on |
An important caveat: NER models make mistakes, especially with unusual names, abbreviations, or industry-specific jargon. Human review of exceptions remains wise, especially for financial or legal applications.
Related concepts
Named entity recognition relates to other concepts in AI and language processing:
- Natural Language Processing (NLP): the broader field named entity recognition falls under.
- Sentiment analysis: determining the emotional tone of text, often combined with NER to see which sentiment relates to which person or organization.
- Data extraction: automatically retrieving structured data from documents, where NER is often the first step.
Want to know how named entity recognition could speed up your document processing, for example through an AI agent that automatically processes invoices or emails? Discuss it in a conversation about AI consultancy, or start with the free AI scan to see where manual entry work in your organization could be automated.
Frequently asked questions
What's the difference between NER and sentiment analysis?
NER recognizes and labels specific elements in text, such as names and dates. Sentiment analysis determines the emotional tone of a text. They're often used together, for example to see which sentiment relates specifically to a mentioned organization.
Can NER also recognize Dutch names and organizations?
Yes, provided the model is trained on Dutch-language text. Models primarily trained on English text often recognize Dutch names and organizations less reliably.
Is named entity recognition the same as OCR?
No. OCR (Optical Character Recognition) converts an image of text (like a scan) into machine-readable text. NER then works on that text and recognizes specific meaningful elements within it. For scanned documents, OCR is typically applied first, followed by NER.
How accurate is named entity recognition?
This varies by model and application. For common entities like dates or amounts, accuracy is generally high. For unusual names, abbreviations, or industry-specific terms, the error rate increases, which is why sample checking remains advisable.
Veelgestelde vragen
Korte, heldere antwoorden die je helpen sneller beslissen.
What's the difference between NER and sentiment analysis?
NER recognizes and labels specific elements in text, such as names and dates. Sentiment analysis determines the emotional tone of a text. They're often used together, for example to see which sentiment relates specifically to a mentioned organization.
Can NER also recognize Dutch names and organizations?
Yes, provided the model is trained on Dutch-language text. Models primarily trained on English text often recognize Dutch names and organizations less reliably.
Is named entity recognition the same as OCR?
No. OCR converts an image of text (like a scan) into machine-readable text. NER then works on that text and recognizes specific meaningful elements within it. For scanned documents, OCR is typically applied first, followed by NER.
How accurate is named entity recognition?
This varies by model and application. For common entities like dates or amounts, accuracy is generally high. For unusual names or industry-specific terms, the error rate increases, which is why sample checking remains advisable.
