Year: 2026
Type: conference-paper
Abstract: Encoding graphs as text prompts enables LLMbased graph reasoning.However, effective graph-to-text encodings depend on the downstream task, graph structure, and target LLM, making manual format design time-consuming and brittle.We propose EVOFORMAT, a blackbox evolutionary search framework that automatically discovers interpretable graph-totext formats represented as human-readable formatting rules over node naming, relation expression, and global arrangement.Using only task-level feedback from LLM outputs, EVOFORMAT iteratively mutates and selects rules without gradients or internal model access.Across synthetic GraphQA and IMDB-MULTIderived structural QA tasks over unattributed graphs, EVOFORMAT outperforms strong handcrafted encodings in most settings.
Institution: NEC (Japan)
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W7224504530