Just to be clear, Torus uses tsvector as well, so you may be comparing apples to the same apples.
christhekeele
1
hakarabakara1
Hello,
Thank you again for this package. I am trying to migrate my previous search which already leveraged upon Postgres triggers to generate tsvector on record create with rankings already depending on column. Should I just use similarity to query the column or do I need to create embedding with the recommended packages to get the best out of Torus?
0
dimamik
Torus v0.7.0 is out - Hybrid Search and Match Highlighting!
Hybrid search via the new Torus.hybrid/4 macro - it combines several search strategies (full_text, bm25, similarity, semantic) into a single ranked query using Reciprocal Rank Fusion, the industry-standard way to combine keyword and semantic search:
search_vector = Torus.to_vector("magic school")
Post
|> Torus.hybrid([p], [
full_text: {[p.title, p.body], "magic school"},
semantic: {p.embedding, search_vector, weight: 2.0}
], limit: 10)
|> Repo.all()
Everything runs in a single SQL query, and the result is a regular Ecto query you can keep piping select, preload, or pagination onto. Per-branch weights and limits are tunable. See the Hybrid search guide for details.
Match highlighting via the new Torus.highlight/3 macro, built on ts_headline. The simplest way is the new :highlight option available on all search types:
Post
|> Torus.full_text([p], [p.title, p.body], "shocker", highlight: [title: p.title])
|> Repo.all()
# => [%Post{title: "Hogwarts <b>Shocker</b>", ...}]
Try both live on the demo page. Let me know what you think!
Heads up on breaking changes: Torus now requires ecto ~> 3.10, and Torus.Embeddings.NebulexCache targets Nebulex 3.0+ (add nebulex_local if you use the cache). Full details in the changelog.
3