xb

Vector Diversity & Qdrant (English)

English remake of xb/doc/VECTOR_DIVERSITY_QDRANT.md. It explains how xb balances recommendation diversity using Qdrant helpers.

---

Diversity strategies

Strategy	Builder helper	Description
Hash diversity	WithHashDiversity(func(*HashDiversity))	Spreads results across payload buckets
Score floor	WithMinDistance(float32)	Drops items below a similarity threshold
Payload projection	WithPayloadSelector	Controls fields returned for downstream rerankers

---

Example

custom := xb.NewQdrantBuilder().
    WithHashDiversity(func(h *xb.HashDiversity) {
        h.Field = "category"
        h.Modulo = 4
    }).
    WithMinDistance(0.35).
    Build()

json, _ := xb.Of(&ProductVector{}).
    Custom(custom).
    VectorSearch("embedding", vec, 12).
    Build().
    JsonOfSelect()

---

Best practices

• Choose hash fields that correlate with business diversity (brand, category, tenant).
• Tune modulo to match the number of slots you want.
• Combine with server-side rerankers if you need deterministic diversity.

---

Related docs

• doc/en/VECTOR_GUIDE.md
• doc/en/QDRANT_OPTIMIZATION_SUMMARY.md

This site is open source. Improve this page.