When LLMs meet open-world graph learning: a new perspective for unlabeled data uncertainty

TL;DR

This paper introduces OGA, an LLM-based framework for open-world graph learning that effectively handles data uncertainty, unknown classes, and limited labels through adaptive traceability and annotation pipelines.

Contribution

The paper presents a novel LLM-driven framework, OGA, combining semantic and structural analysis for unknown-class rejection and enabling dynamic model updates with new annotations.

Findings

OGA outperforms existing methods in unknown-class rejection.

OGA effectively integrates semantics and topology for better uncertainty handling.

Experimental results validate OGA's practicality and robustness.

Abstract

Recently, large language models (LLMs) have significantly advanced text-attributed graph (TAG) learning. However, existing methods inadequately handle data uncertainty in open-world scenarios, especially concerning limited labeling and unknown-class nodes. Prior solutions typically rely on isolated semantic or structural approaches for unknown-class rejection, lacking effective annotation pipelines. To address these limitations, we propose Open-world Graph Assistant (OGA), an LLM-based framework that combines adaptive label traceability, which integrates semantics and topology for unknown-class rejection, and a graph label annotator to enable model updates using newly annotated nodes. Comprehensive experiments demonstrate OGA's effectiveness and practicality.