Can LLM Graph Reasoning Generalize beyond Pattern Memorization?

arXiv:2406.15992v1 Announce Type: new
Abstract: Large language models (LLMs) demonstrate great potential for problems with implicit graphical structures, while recent works seek to enhance the graph reasoning capabilities of LLMs through specialized instruction tuning. The resulting 'graph LLMs' are evaluated with in-distribution settings only, thus it remains underexplored whether LLMs are learning generalizable graph reasoning skills or merely memorizing patterns in the synthetic training data. To this end, we propose the NLGift benchmark, an evaluation suite of LLM graph reasoning …

abstract arxiv beyond capabilities cs.cl distribution graph instruction tuning language language models large language large language models llm llms pattern potential reasoning seek the graph through tuning type while