NLP (Concepts & Pipeline)
In shortNamed entity recognition: finding names of people, organisations, places, products and other things in text and labelling each with a type.
Entity extraction, known in research as named entity recognition (NER), finds the names of real-world things in text, such as people, organisations, places, products, events and dates, and labels each with a type. It turns unstructured text into structured data, which is why it underpins entity SEO, content audits and much of how search engines and AI systems understand a page.
NER is a supervised, prediction-based task. A model is trained on texts where entities are already highlighted (for example news articles with every person’s name marked), learns the patterns, and then tags entities in text it has never seen. Within the NLP pipeline it belongs to the semantic analysis phase, alongside sentiment analysis and topic modelling.
Pre-trained APIs return more than names. The Google Natural Language API adds the entity type, a salience score, mentions, sentiment and, where available, a Knowledge Graph ID and Wikipedia URL. That link to a knowledge base is called entity linking, and it is what lets you check which real-world thing a name refers to. Systems are evaluated with precision, recall and F1, a standard set by the CoNLL-2003 shared task.
In the MLforSEO benchmark, Google’s API returned 59 unique entities from a 221-word bio and found all 13 core named entities, but 31 of the 59 were generic nouns like “teams” that you filter out using salience and type. IBM Watson linked “AWS” to the American Welding Society rather than Amazon Web Services, a reminder to check knowledge-base links on short acronyms.
