IGDA: Interactive Graph Discovery through Large Language Model Agents

ACL ARR 2025 May Submission3310 Authors

Abstract: Large language models have emerged as a powerful tool for accelerating science and decision making. Towards further improving LLM utility in these domains we study the application of LLMs to the novel task of \textit{interactive graph discovery}: given a ground truth graph $G^*$ capturing variable relationships and a budget of $I$ edge experiments over $R$ rounds, minimize the distance between the predicted graph $\hat{G}_R$ and $G^*$ at the end of the $R$-th round. To solve this task we propose \textbf{IGDA}, a LLM-based pipeline incorporating two key components: 1) an LLM uncertainty-driven method for edge experiment selection 2) a local graph update strategy utilizing binary feedback from experiments to improve predictions for unselected neighboring edges. Experiments on eight different real-world graphs show our approach often outperforms all baselines including a state-of-the-art numerical method for interactive graph discovery. Further, we conduct a rigorous series of ablations dissecting the impact of each pipeline component. Overall, our results show IGDA to be a powerful method for graph discovery complementary to existing numerically driven approaches.

Paper Type: Long

Research Area: Language Modeling

Research Area Keywords: LLMs,Discovery,Uncertainty,Causal

Contribution Types: NLP engineering experiment

Languages Studied: English

Submission Number: 3310