From Injection to Interaction: A Survey on Knowledge Graph-Augmented LLM Reasoning

The Hong Kong University of Science and Technology (Guangzhou)

Data Science and Analytics Thrust

PhD Qualifying Examination

By Miss ZHAO, Wenxin

Abstract

In recent years, Large Language Models (LLMs) have achieved remarkable success across a wide range of natural language understanding and generation tasks. Despite their impressive capabilities, LLMs often struggle with factual reliability and logical consistency, particularly in knowledge-intensive settings. This is largely due to their reliance on parametric memory, which cannot be easily updated or externally verified once training is complete. As a result, LLMs are prone to hallucinations and limited in their ability to reason over structured, domain-specific information. These challenges underscore the growing need to integrate LLMs with external symbolic resources, such as knowledge graphs (KGs), which offer explicit, structured representations of real-world facts. This survey provides a comprehensive overview of KG-enhanced LLM methods, highlighting recent progress, categorizing mainstream approaches, and identifying open research challenges. Through asystematic comparative analysis, it uncovers new insights and offers a fresh perspective on the integration of LLMs and knowledge graphs, laying the groundwork for future research into more reliable, interpretable, and knowledge-aware AI systems.

PQE Committee

Chair of Committee: Prof. TANG Nan

Prime Supervisor: Prof. CHEN Lei

Co-Supervisor: Prof. ZHANG Yongqi

Examiner: Prof. ZHANG Yanlin