ACORN Algorithm

Overview

ACORN (Approximate k-Nearest Neighbors with HNSW Over Refined Neighborhoods) is an approach for performant and predicate-agnostic hybrid search that builds on HNSW and can be implemented efficiently by extending existing HNSW libraries.

Two-Hop Neighborhood Strategy

The two-hop expansion is a key innovation in ACORN:

• Evaluates nodes that are two hops away rather than one
• Target nodes become out-neighbors of search nodes (two hops away)
• Speeds up graph traversal
• Ameliorates challenges of low correlation between filters and query vectors

How It Works

ACORN builds on Hierarchical Navigable Small Worlds (HNSW), a state-of-the-art graph-based approximate nearest neighbor index. The algorithm uses a two-hop based expansion of the neighborhood to maintain search quality when filters are applied.

Weaviate Implementation (2026)

Weaviate uses ACORN to keep recall high under selective filters:

Conditional Two-Hop Evaluation

• If the first hop passes the filter, traverse graph normally
• Only use two-hop expansion if connecting node doesn't pass filter
• Optimizes performance based on filter selectivity

Benefits in Production

• Combines semantic similarity with structured filters
• Keeps filtered queries efficient on realistic datasets
• Designed specifically for vector search with rich filtering over high-dimensional spaces

Performance Characteristics

• Maintains high recall even with selective filters
• Efficient implementation by extending HNSW libraries
• Predicate-agnostic: works well regardless of filter characteristics

Research Impact

Published in Proceedings of the ACM on Management of Data (2024), ACORN represents a significant advancement in filtered vector search, addressing one of the main challenges in production vector database deployments.

Applications

• E-commerce search with category/price filters
• Document retrieval with metadata constraints
• Multi-tenant vector databases
• Any scenario requiring hybrid semantic + structured search