Keyword and vector search

At a glance

Keyword signal

Text terms and their matching rules

Vector signal

Similarity between embeddings

Overview

Keyword search uses terms in a query to find matching text, while vector search compares numeric representations produced by an embedding model. Each exposes different kinds of similarity, and systems can combine them when exact terminology and broader semantic relationships both matter.

What differs

Keyword retrieval can preserve the importance of a specific name or phrase. Embedding-based retrieval can find related phrasing even when words differ, but its ranking depends on the model and similarity method.

Choose by the task

Evaluate realistic queries and known relevant results. Product identifiers, specialized vocabulary, and broad natural-language questions may benefit from different signals or a combined retrieval approach.

Common confusion

A high similarity score does not prove that a result answers the question. Check relevance against the user's task and retain source references so the retrieved material can be inspected.

Sources and review

MOOR's explanatory text is supported by the following source links.

1. Introduction to full-text search — PostgreSQL
2. Embeddings — Cohere