Reproduce “LLaTA: Unlocking Graph Structure Learning with Tree-Guided Large Language Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhihan Zhang, Xunkai Li, Hongchao Qin, Rong-Hua Li, Guoren Wang
LLaTA improves text-attributed graph structure by building an entropy-based hierarchy of communities and using an LLM’s class-conditioned text judgments to guide targeted edge edits, without fine-tuning. The learned structure is evaluated across 11 datasets with downstream encoders. On Citeseer node classification, LLaTA reports 78.21 ± 0.18% accuracy versus 76.96 ± 0.58% for LLM4RGNN, a 1.25-point difference; this result is task- and label-conditioned.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.