Graph paper: a curated map of LLM research on networks

What it does

This is a curated bibliography—an “awesome list”—that catalogs papers exploring how large language models handle graph data. It is organized around a taxonomy from the authors’ own TKDE survey, splitting work into pure graphs, text-attributed graphs, and text-paired graphs such as molecules. Each entry is tagged with architectural details like DecoderOnly and model size, turning a sprawling literature into something browsable.

The interesting bit

The taxonomy itself is the product. Rather than dumping links, the authors impose order on a messy convergence of two fields, distinguishing between tasks like direct graph answering, heuristic reasoning, and algorithmic reasoning. They also track whether a work treats the LLM as a predictor, encoder, or aligner, which reveals how researchers are currently trying to bridge sequential text and structured topology.

Key highlights

• Organizes papers across three graph types: pure graphs, text-attributed graphs, and text-paired molecular graphs.
• Tags entries with model architecture and scale (medium vs. LLM) so you can filter by computational heft.
• Covers distinct interaction modes: LLM as predictor, encoder, and aligner.
• Includes dedicated dataset sections with tables from the underlying survey paper.
• Maintains a sister repository for multimodal learning on graphs.

Caveats

• This is a reading guide, not a framework or codebase; if you are looking for libraries to import, this is not it.
• Many entries are marked as preprint, so the field is still shifting under the taxonomy.
• The README promises continuous updates, which is standard for awesome lists but means some sections may still be filling out.

Verdict

Grab this if you are a researcher or practitioner trying to navigate the LLM-on-graph literature without drowning in arXiv alerts. Skip it if you need a drop-in library to run graph reasoning today.