Ask Safely: Privacy-Aware LLM Query Generation for Knowledge Graphs

2025-12-04

AI Summary

To address the challenge of securely querying knowledge graphs containing sensitive data using third-party large language models (LLMs) under resource-constrained conditions, this paper proposes a privacy-preserving natural language-to-Cypher query generation method. The approach automatically identifies and filters sensitive nodes via graph-structure analysis, performing lightweight input-side anonymization without requiring local LLM deployment—enabling safe invocation of cloud-based LLM services. Its core innovation lies in jointly modeling sensitive information identification, structure-aware anonymization, and LLM-driven semantic translation to simultaneously ensure privacy preservation and query accuracy. Experiments on multiple benchmark datasets demonstrate that the method maintains over 90% Cypher generation accuracy while strictly preventing leakage of sensitive entities and relationships, significantly outperforming baseline anonymization strategies.

Technology Category

• Data Mining & Knowledge Management: Linked Open Data, Knowledge Graphs & KB Completion
• Machine Learning: Privacy
• Natural Language Processing: (Large) Language Models

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Large language models for search

Abstract

Large Language Models (LLMs) are increasingly used to query knowledge graphs (KGs) due to their strong semantic understanding and extrapolation capabilities compared to traditional approaches. However, these methods cannot be applied when the KG contains sensitive data and the user lacks the resources to deploy a local generative LLM. To address this issue, we propose a privacy-aware query generation approach for KGs. Our method identifies sensitive information in the graph based on its structure and omits such values before requesting the LLM to translate natural language questions into Cypher queries. Experimental results show that our approach preserves the quality of the generated queries while preventing sensitive data from being transmitted to third-party services.

Problem

• Prevents sensitive data exposure in KG queries
• Generates queries without local LLM deployment
• Maintains query quality while ensuring privacy

Innovation

• Privacy-aware query generation for knowledge graphs
• Identifies sensitive data based on graph structure
• Omits sensitive values before LLM query translation