When LLMs Encounter Open-world Graph Learning: A Fresh View on Unlabeled Data Uncertainty
Abstract
Recently, large language models (LLMs) have driven a systematic shift in the graph ML community through the adoption of text-attributed graphs (TAGs). Although a variety of frameworks have been developed, most fail to properly address the challenge of data uncertainty in openworld environments. A representative source of such uncertainty is the limited availability of labels in large-scale datasets due to high annotation costs, where unlabeled nodes may belong to either known classes or novel, unknown classes. While node-level out-of-distribution detection and conventional open-world graph learning attempt to tackle this problem, two core limitations remain: (1) Insufficient methods — existing approaches typically optimize semantics or topology in isolation for unknown-class rejection, failing to effectively integrate textual and structural information in TAGs; (2) Incomplete pipelines — most studies conduct only idealized analyses, such as assuming a predefined number of unknown classes, which restricts practical utility for model updates and long-term deployment. To overcome these issues, we introduce the Open-world Graph Assistant (OGA), an LLM-based framework. OGA first performs unknown-class rejection via adaptive label traceability (ALT), harmoniously combining semantic and topological cues, and then applies the graph label annotator (GLA) for unknownclass annotation, allowing unlabeled nodes to contribute to model training. In essence, OGA offers a new pipeline that fully automates the handling of unlabeled nodes in open-world environments, and we establish a systematic benchmark covering four key aspects to validate its effectiveness and practicality through extensive experiments.