starred/minne
1
0
Fork 0
mirror of https://github.com/perstarkse/minne.git synced 2026-06-21 14:09:33 +02:00
Code Issues Packages Projects Releases 10 Wiki Activity

fix: knowledge entity suggestions simplification

Browse Source
This commit is contained in:
Per Stark
2026-05-31 20:23:40 +02:00
parent 3897345ab3
commit b22c351785
9 changed files with 394 additions and 122 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/features.md
+7 -5
View File
@@ -27,14 +27,16 @@ The D3-based graph visualization shows entities as nodes and relationships as ed
## Hybrid Retrieval
Minne uses chunk-first hybrid retrieval over the knowledge base:
Minne uses hybrid retrieval over the knowledge base:
- **Vector similarity** — Semantic matching via embeddings over text chunks
- **Full-text search** — Keyword matching with BM25 over the same chunk index
- **Vector similarity** — Semantic matching via embeddings
- **Full-text search** — Keyword matching with BM25
The two ranked candidate lists are merged with Reciprocal Rank Fusion (RRF). When a caller needs knowledge entities (search, ingestion linking, relationship suggestion), entities are derived from the top retrieved chunks grouped by `source_id`.
For **content search** (chat, global search, ingestion linking), retrieval is chunk-first: vector and FTS run over `text_chunk` rows, merged with Reciprocal Rank Fusion (RRF). When entities are needed, they are derived from the top retrieved chunks grouped by `source_id`.
Optional **reranking** can rescore the fused chunk list with a cross-encoder model; see below.
For **relationship suggestions** when creating an entity, retrieval is entity-first: vector and FTS run directly over `knowledge_entity` name/description and embedding indexes, then merged with the same RRF approach.
Optional **reranking** can rescore fused chunk lists with a cross-encoder model; see below.
## Reranking (Optional)