Entity Extraction

Entity extraction is a natural language processing technique that identifies and classifies named entities and other structured information within unstructured text. Entities typically include person names, organisations, locations, dates, monetary amounts, product names, and domain-specific terms. The technology has evolved from rule-based systems and statistical models to modern transformer-based approaches and LLM-powered extraction. Traditional Named Entity Recognition (NER) models identify standard entity types. LLM-based extraction can handle arbitrary entity types defined through prompts, making it far more flexible. Entity extraction can operate at different levels of specificity. Basic extraction identifies entity mentions and their types. More advanced systems resolve entities (linking 'Apple' to the correct company or fruit), extract relationships between entities (identifying that a person works for a company), and normalise values (converting various date formats to a standard form).

Entity extraction transforms unstructured documents into structured data that can be searched, analysed, and acted upon. Legal teams use it to extract parties, dates, and obligations from contracts. Financial teams extract amounts, counterparties, and transaction types from communications. Healthcare systems extract diagnoses, medications, and procedures from clinical notes. When combined with knowledge graphs, extracted entities create rich, interconnected data models that enable sophisticated queries and analyses. An organisation can map relationships between customers, products, issues, and resolutions across millions of documents, revealing patterns invisible to manual analysis. Modern LLMs have made entity extraction significantly more accessible. Rather than training specialised NER models for each entity type, organisations can use general-purpose LLMs with appropriate prompts to extract arbitrary information from text, dramatically reducing the time and expertise needed to deploy extraction pipelines.