4 ms·
Wouldn't holy grail then be parallel channels for candidate generation; euclidean embedding hyperbolic embedding sparse BM25 / SPLADE lexical search op
by tkfoss 1y ago
Wouldn't holy grail then be parallel channels for candidate generation;
euclidean embedding
hyperbolic embedding
sparse BM25 / SPLADE lexical search
optional multi-vector signatures
↓ merge & deduplicate candidates
followed by weight scoring, expansion (graph) & rerank (LLM)?
- jdthedisciple 1y agothat is pretty much exactly what we do for our company-internal knowledge retrieval: embedding search (0.4) lexical/keyword search (0.4) fuzzy search (0.2) might indeed achieve the best of all worlds