Entities are the elements named in a text that correspond to something in the real world. People, organisations, places, products, dates and monetary amounts all fit the definition. What separates them from the rest of the text is that they point at something specific: "Istanbul" is a particular city, "14 March 2024" a particular day.
Extracting entities is one of the core jobs in NLP. Named entity recognition models find the candidates in text and label them by type. Entity linking comes next, mapping each one to a unique identifier in a knowledge base. That is how "Apple" and "Apple Inc." arrive at the same record and the system knows the two spellings mean one thing.
On the search side entities matter more every year. Search engines assess a page not only by its words but by the entities inside it and the relationships between them. Naming the organisations, products and places on a page clearly makes it easier for a machine to place the content in a subject.
A concrete case: a travel site publishes a page on hot air balloon tour prices in Cappadocia. Cappadocia is a place entity in that text and the balloon tour is a product category. Marking those entities up with structured data too makes the page match the relevant query set more cleanly.
Ambiguity never goes away. One name can correspond to several entities, so linking to the right identifier is impossible without context.
