[Paper Review] Navigable Proximity Graph-Driven Native Hybrid Queries with Structured and Unstructured Constraints

Mengzhao Wang, Lingwei Lv+4 more|arXiv (Cornell University)|Mar 25, 2022

Advanced Image and Video Retrieval Techniques4 citations

TL;DR

This paper proposes a native hybrid query (NHQ) framework built on navigable proximity graphs (NPGs) that jointly optimizes vector similarity search and structured attribute filtering for hybrid queries. By integrating specialized composite indexing and joint pruning, the NHQ framework achieves up to 10× faster performance than state-of-the-art methods while maintaining high recall across eight public and one in-house dataset.

ABSTRACT

As research interest surges, vector similarity search is applied in multiple fields, including data mining, computer vision, and information retrieval. {Given a set of objects (e.g., a set of images) and a query object, we can easily transform each object into a feature vector and apply the vector similarity search to retrieve the most similar objects. However, the original vector similarity search cannot well support extit{hybrid queries}, where users not only input unstructured query constraint (i.e., the feature vector of query object) but also structured query constraint (i.e., the desired attributes of interest). Hybrid query processing aims at identifying these objects with similar feature vectors to query object and satisfying the given attribute constraints. Recent efforts have attempted to answer a hybrid query by performing attribute filtering and vector similarity search separately and then merging the results later, which limits efficiency and accuracy because they are not purpose-built for hybrid queries.} In this paper, we propose a native hybrid query (NHQ) framework based on proximity graph (PG), which provides the specialized extit{composite index and joint pruning} modules for hybrid queries. We easily deploy existing various PGs on this framework to process hybrid queries efficiently. Moreover, we present two novel navigable PGs (NPGs) with optimized edge selection and routing strategies, which obtain better overall performance than existing PGs. After that, we deploy the proposed NPGs in NHQ to form two hybrid query methods, which significantly outperform the state-of-the-art competitors on all experimental datasets (10$ imes$ faster under the same extit{Recall}), including eight public and one in-house real-world datasets. Our code and datasets have been released at \url{https://github.com/AshenOn3/NHQ}.

Motivation & Objective

• To address the inefficiency and inaccuracy of existing hybrid query processing methods that treat vector similarity search and attribute filtering as separate steps.
• To design a unified framework that natively supports both unstructured (feature vector) and structured (attribute) constraints in a single query execution.
• To develop optimized navigable proximity graphs (NPGs) with improved edge selection and routing strategies for better performance in hybrid query workloads.
• To demonstrate significant performance gains over existing state-of-the-art systems in real-world and public datasets.

Proposed method

• Proposes a native hybrid query (NHQ) framework that integrates vector similarity search and attribute filtering into a single, joint processing pipeline using proximity graph (PG) structures.
• Introduces a composite index structure that combines feature vectors and attribute metadata within the PG for efficient hybrid query execution.
• Designs joint pruning modules that simultaneously reduce the search space for both vector similarity and attribute constraints during traversal.
• Develops two novel navigable PGs (NPGs) with optimized edge selection and routing strategies tailored for hybrid queries, improving overall retrieval performance.
• Deploys the proposed NPGs within the NHQ framework to create two hybrid query methods: NHQ-NPG_kgraph and NHQ-NPG_sprawl.
• Employs a dynamic weight adjustment mechanism in the routing strategy to balance vector similarity and attribute constraint satisfaction.

Experimental results

Research questions

• RQ1 Can a unified framework be designed to natively support hybrid queries combining unstructured and structured constraints, avoiding the inefficiencies of sequential processing?
• RQ2 How can proximity graphs be enhanced with composite indexing and joint pruning to improve hybrid query performance?
• RQ3 What are the optimal edge selection and routing strategies for navigable proximity graphs in hybrid query scenarios?
• RQ4 Can the proposed NHQ framework achieve significant performance gains over existing state-of-the-art hybrid query systems?

Key findings

• The NHQ framework achieves up to 10× speedup over state-of-the-art competitors while maintaining the same recall level across all evaluated datasets.
• The proposed NHQ-NPG_kgraph and NHQ-NPG_sprawl methods outperform existing systems on all eight public and one in-house real-world datasets.
• The optimized edge selection and routing strategies in the two novel NPGs lead to better overall performance compared to existing PGs.
• The use of joint pruning and composite indexing reduces redundant search space and improves query efficiency without sacrificing accuracy.
• The framework is compatible with existing PGs, enabling easy deployment of various PGs for hybrid query processing.
• A case study in an academic expert-finding system shows that NHQ-NPG_kgraph retrieves more accurate and relevant experts than HNSW and Vearch, especially when attribute constraints are involved.