Towards the Effect of Large Language Models on Out-Of-Distribution Challenge in Text-Attributed Graphs
Keywords: Out-Of-Distribution, Large Language Models, Text-Attributed-Graphs
Abstract: Text-Attributed Graphs (TAGs), where each node is associated with text attributes, are ubiquitous and have been widely applied in the real world. The Out-Of-Distribution (OOD) issue, i.e., the training data and the test data not from the same distribution, is quite common in learning on real-world TAGs, posing significant challenges to the effectiveness of graph learning models. Recently, Large Language Models (LLMs) have shown extraordinary capability in processing text data, and have demonstrated tremendous potential in handling TAGs. However, there is no benchmark work that systematically and comprehensively investigates the effect of these LLM-based methods on alleviating the OOD issue on TAGs. To bridge this gap, we first develop OOD-TAG, a comprehensive OOD benchmark dataset in TAGs which consists of diverse distributions. Meanwhile, we conduct a systematic and comprehensive investigation on OOD-TAG with different LLM pipelines for graphs. In addition, we provide original observations and novel insights based on the empirical study, which can suggest promising directions for the research of LLMs in addressing the OOD challenges on TAGs. Our code and dataset are available in https://anonymous.4open.science/r/GraphOOD-benchmark-5FCF/.
Primary Area: datasets and benchmarks
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 158
Loading