A model can represent text or other content as vectors, which are ordered lists of numbers. A search system compares a query vector with stored vectors to retrieve similar items.
Similarity is not the same as truth or a structured relationship. Vector search can complement exact filters and graph queries, especially when wording varies across documents.
ELI5
Vector search helps find material that is similar in meaning even when it uses different words. It turns each item into a numerical representation that can be compared.
For example, a search for ways to lower a heating bill might retrieve a guide titled saving energy at home. Similarity helps find it, but the system still needs to check whether the guide answers the question.
Is similarity proof of relevance?
No. Similar items can be unhelpful or misleading, so retrieval results still need evaluation.
Can it be combined with exact search?
Yes. Exact filters and other retrieval methods can narrow results or supply structure that similarity alone lacks.
