FairGT: A Fairness-aware Graph Transformer

Authors: Renqiang Luo, Huafei Huang, Shuo Yu, Xiuzhen Zhang, Feng Xia

IJCAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirical evaluations conducted across five real-world datasets demonstrate Fair GT s superiority in fairness metrics over existing graph transformers, graph neural networks, and state-of-theart fairness-aware graph learning approaches.
Researcher Affiliation Academia 1Dalian University of Technology, China 2RMIT University, Australia
Pseudocode No The paper does not contain any sections explicitly labeled 'Pseudocode' or 'Algorithm', nor are there structured, code-like blocks detailing a procedure.
Open Source Code Yes 1The source codes and detailed proofs are available at https://github.com/yushuowiki/Fair GT.
Open Datasets Yes In this study, node classification serves as the downstream task, employing five distinct real-world datasets: NBA, Bail, German, Credit, and Income. The statistics of five datasets are shown in Table 3. NBA [Dai and Wang, 2021]: The NBA dataset, sourced from Kaggle... Bail [Jordan and Freiburger, 2015]: This dataset represents defendants... German [Asuncion and Newman, 2007]: German is extracted from the Adult Data Set. ... Credit [I-cheng and Che-hui, 2009]: The Credit dataset uses personal next month. ... Income [Asuncion and Newman, 2007]: Income is extracted from the Adult Data Set.
Dataset Splits Yes Then, we randomly select 25% nodes as the validation set and 25% as the test set, ensuring a balanced ratio of ground truth labels.
Hardware Specification No The paper does not explicitly specify any hardware details (e.g., specific GPU or CPU models, memory) used for running the experiments. It only discusses training time comparisons.
Software Dependencies No The paper does not provide specific version numbers for any software dependencies or libraries used in the experimental setup.
Experiment Setup Yes For a fair comparison, we standardize key parameters across all methods, setting the number of hidden dimensions to 128, the number of layers to 1, the number of heads to 1, and the number of epochs to 500. ...We select t = 5 for NBA, t = 5 for Bail, t = 11 German, t = 2 for Credit, and t = 2 for Income. ...We set l = 2 for NBA, l = 3 for Bail, l = 3 for German, l = 2 for Credit, and l = 1 for Income.