🔍 Retrieval methods, side by side

Compare BM25 (lexical), Dense (vector cosine), Hybrid (Reciprocal Rank Fusion), and an optional cross-encoder rerank over a corpus of 3,993 Europe PMC abstracts on microRNA & disease.

Query

Top-k

1 10

Embedding model — used by Dense, Hybrid and the plot (BM25 is lexical, so it never changes)

General — MiniLM-L6 (384-dim) Biomedical — S-PubMedBert (768-dim)

Metadata filter — restricts the candidate pool before retrieval (all methods).

Year from

2023 2026

Year to

2023 2026

Journal

—

Click any result to expand its full abstract.

Vector space (UMAP projection)

Plot

⚠️ The 2D UMAP projection distorts true distances. UMAP preserves rough local neighbourhoods but warps global distances, and the live query is placed by an approximate out-of-sample fit — so its position (and the connector lines) are only indicative. Retrieved points may not be the visually-closest dots. The ranked lists above are authoritative; the plot is only for intuition.