Both Topology and Text Matter: Revisiting LLM-guided Out-of-Distribution Detection on Text-attributed Graphs
Researchers have developed LG-Plug, a novel framework designed to improve out-of-distribution (OOD) detection in text-attributed graphs (TAGs). This method effectively combines graph topology with textual information, addressing limitations of previous approaches that either underutilized semantic data or struggled with reliable OOD exposure generation. LG-Plug integrates seamlessly with existing detectors, significantly enhancing their performance and reducing false positives on unseen data. AI
IMPACT Enhances the reliability of AI models in identifying novel or out-of-distribution data, crucial for robust real-world applications.