LLM+KG@VLDB'24 Workshop Summary

TL;DR

Problem addressed: LLMs risk hallucinations without grounding, and KGs provide factual grounding but require effective integration. The workshop surveys three convergence themes—LLMs for KG construction and QA, KG grounding for LLMs, and unified LLM+KG systems—covering both research and industry perspectives, including GraphRAG and SPG platforms. Key contributions include KnowLA for KG-enhanced PEFT with LoRA, SPG's layered architecture, GraphRAG retrieval strategies, and neural-symbolic knowledge editing (OneEdit) and SPIREX-based triple extraction, plus panel-driven guidance on future directions. The findings highlight practical paths for scalable, auditable, and privacy-conscious data-management workflows in AI systems and stress the need for benchmarks, open resources, and academia-industry collaboration.

Abstract

The unification of large language models (LLMs) and knowledge graphs (KGs) has emerged as a hot topic. At the LLM+KG'24 workshop, held in conjunction with VLDB 2024 in Guangzhou, China, one of the key themes explored was important data management challenges and opportunities due to the effective interaction between LLMs and KGs. This report outlines the major directions and approaches presented by various speakers during the LLM+KG'24 workshop.