8 unstable releases (3 breaking)
0.4.0 | May 3, 2024 |
---|---|
0.3.0 | Dec 10, 2023 |
0.2.2 | Apr 28, 2023 |
0.2.1 | Mar 25, 2023 |
0.1.1 | May 29, 2022 |
#42 in Database implementations
535KB
11K
SLoC
searchy: an embeddable in-memory search engine
Search index (in-memory), that can be constructed and searched using a bag of words model, with cosine similarity scoring based on tf-idf. Supports multiple search terms, permissions, text suggestions (for spelling errors), and evaluating common arithmetic expressions directly to values. Tries to reduce memory footprint.
Features:
- expression evaluation (1+2 will result in 3);
- small in memory representation using two allocations (after building the index);
- easy delta updates;
- spell suggestions based on search index;
- summary sentence in search results (based on Luhn);
- filtering of result based on group (role-based access control).
Minimal example
use searchy::*;
use expry::*;
// MemoryPool avoids small allocations.
pool!(scope);
// Add documents
let mut builder = SearchBuilder::new();
let doc_id = builder.add_document("foo bar text", "url", "name", "group1", b"extra", &mut scope);
// doc_id can be used to later remove the document from the search index
// Build search index
let index : SearchIndex = builder.into();
// Roles are access groups of the user. The search documents have access groups and are are filtered based on the access group of the current user.
const MAX_DOCS_RETRIEVED : usize = 1024;
let results = index.search("query text", "group1,group2", MAX_DOCS_RETRIEVED);
eprintln!("RESULTS: {}x/{} in {}ms", results.docs, results.more, results.duration);
for ScoredDocumentInfo{doc_id: _, score, info} in results.entries {
eprintln!("{} -> {}, {}, {}", score, info.name, info.url, info.summary);
}
// do some mutations to the search index
let mut builder = SearchBuilder::from(index);
builder.remove_document(doc_id);