Position-aware Graph Neural Networks

Authors: Jiaxuan You, Rex Ying, Jure Leskovec

ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We apply P-GNNs to 8 different datasets and several different prediction tasks including link prediction and community detection1. In all datasets and prediction tasks, we show that P-GNNs consistently outperforms state of the art GNN variants, with up to 66% AUC score improvement. (Section 6: Experiments, Table 1 and Table 2 present numerical results from these experiments)
Researcher Affiliation Academia 1Department of Computer Science, Stanford University, Stanford, CA, USA.
Pseudocode Yes Algorithm 1 The framework of P-GNNs
Open Source Code Yes Code and data are available in http://snap.stanford.edu/pgnn/
Open Datasets Yes PPI. 24 Protein-protein interaction networks (Zitnik & Leskovec, 2017). Emails. 7 real-world email communication graphs from SNAP (Leskovec et al., 2007)
Dataset Splits Yes Specifically, we follow the experimental setting from (Zhang & Chen, 2018), and use two sets of 10% existing links and an equal number of nonexistent links as test and validation sets, with the remaining 80% existing links and equal number of nonexistent links used as the training set.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU/GPU models, memory, or cloud instance types) used for running its experiments.
Software Dependencies No The paper does not provide specific software dependencies with version numbers.
Experiment Setup No The paper mentions that 'Implementational details are provide in the Appendix.' but does not include specific experimental setup details like hyperparameter values in the main text.