A roadmap for the awkward union of LLMs and knowledge graphs

What it does

This repository is a curated bibliography of papers exploring how large language models and knowledge graphs can compensate for each other’s weaknesses. LLMs often fall short of capturing factual knowledge, while knowledge graphs are hard to construct and struggle with incomplete or dynamically changing data. The repo collects research that tries to bridge that gap, organized into a three-part taxonomy: using KGs to improve LLMs, using LLMs to improve KGs, and approaches where the two are mutually reinforced.

The interesting bit

The maintainers are active researchers in this space, and the list doubles as a bibliography for their own IEEE TKDE roadmap paper. That gives the taxonomy some academic weight, though it also means the “News” section prominently features their own ICLR, ICML, and NeurIPS papers alongside the broader reading list.

Key highlights

• Covers the full pipeline from pre-training (ERNIE, K-BERT) to inference-time retrieval and interpretability.
• Organizes papers into three frameworks: KG-enhanced LLMs, LLM-augmented KGs, and synergized approaches.
• Includes application areas like recommendation and fault analysis, not just core reasoning.
• Anchored by a published survey paper with a citation-ready BibTeX entry.
• Maintained by authors who also publish in the area, so the list is fed by active research.

Caveats

• It is a reading list, not a toolkit: expect markdown and arXiv links, not installable code.
• The README explicitly notes the project is still under development, so coverage may shift.
• The maintainers’ own work is heavily featured in the news and citations, which frames the perspective.

Verdict

Bookmark this if you are trying to navigate the LLM-plus-KG literature without drowning in arXiv alerts. Skip it if you are hunting for a drop-in library or SDK.