Named Entity Recognition (Ner)

Named Entity Recognition (NER) uses AI to identify and classify entities in text — detecting names of people, organizations, locations, dates, and other entities, providing the foundation for information extraction, knowledge graphs, and semantic understanding.

What Is Named Entity Recognition?

• Definition: Identify and classify named entities in text.
• Entities: People, organizations, locations, dates, products, events, etc.
• Output: Text with entity spans and types labeled.

Common Entity Types

PERSON: Names of people (John Smith, Marie Curie).
ORGANIZATION: Companies, institutions (Apple, MIT, UN).
LOCATION: Cities, countries, landmarks (Paris, USA, Eiffel Tower).
DATE: Dates and times (January 1, 2024, yesterday).
MONEY: Monetary amounts ($100, €50).
PERCENT: Percentages (25%, half).
PRODUCT: Product names (iPhone, Windows).
EVENT: Named events (World War II, Olympics).

Why NER Matters?

• Information Extraction: Extract structured data from text.
• Question Answering: "Who founded Apple?" — need to recognize "Apple" as organization.
• Knowledge Graphs: Populate knowledge bases with entities.
• Search: Entity-aware search and filtering.
• Summarization: Focus on important entities.
• Relation Extraction: Identify relationships between entities.

NER Approaches

Rule-Based: Patterns, gazetteers, regular expressions.
Machine Learning: CRF, SVM with hand-crafted features.
Deep Learning: BiLSTM-CRF, transformers (BERT, RoBERTa).
Transfer Learning: Pre-trained models fine-tuned on NER.
Few-Shot: Learn new entity types from few examples.

Challenges

Ambiguity: "Apple" (company or fruit), "Washington" (person, city, state).
Nested Entities: "Bank of America" contains "America".
Rare Entities: Long-tail entities not in training data.
Domain-Specific: Medical, legal, scientific entities.
Multilingual: Different languages, scripts, naming conventions.

Evaluation Metrics: Precision, recall, F1-score at entity level (exact match or partial match).

Applications: News analysis, customer feedback analysis, legal document processing, medical records, social media monitoring, search engines.

Tools & Models

• Libraries: spaCy, Stanford NER, NLTK, Flair, AllenNLP.
• Models: BERT-NER, RoBERTa-NER, SpanBERT, LUKE (entity-aware).
• Cloud: Google Cloud NLP, AWS Comprehend, Azure Text Analytics.
• Multilingual: mBERT, XLM-R for cross-lingual NER.

Named Entity Recognition is fundamental to NLP — by identifying entities in text, NER enables information extraction, knowledge construction, and semantic understanding, serving as the foundation for countless downstream applications.