✳flâneur — a map of the web's best reading

Semantic Search Without Embeddings

softwaredoug.com · 3,293 words · saved by 1 readers

When I need to stay warm, I search for “long johns” or “long underwear”. But modern outdoors clothing stores label this a “base layer”. For a waterproof jacket I search for “slickers” or a “ski jacket”, not realizing what I should search for is a “shell.” Despite my outdated terminology, search still works. I’m somehow understood and shown the right content. We call this semantic search. When you hear that, you might think embeddings. Today I want to stretch your thinking beyond. Fantasize and semanticize. Some teams wrongly assume semantic search equates to only “vector search”. You have more options, some that might better align to your domain and existing organizational skills. LLMs simplify classic ways of organizing search queries and information. Let’s contrast different approaches, and maybe you’ll see there’s another approach that might be a better fit for you. In semantic search, content and query map to a shared representation. This space has a similarity function that scores

When I need to stay warm, I search for “long johns” or “long underwear”. But modern outdoors clothing stores label this a “base layer”. For a waterproof jacket I search for “slickers” or a “ski jacket”, not realizing what I should search for is a “shell.” Despite my outdated terminology, search still works. I’m somehow understood and shown the right content. We call this semantic search . When you hear that, you might think embeddings. Today I want to stretch your thinking beyond. Fantasize and semanticize. Some teams wrongly assume semantic search equates to only “vector search”. You have mor

Explore this link on the map →

related reading