CLEAR: Generative Counterfactual Explanations on Graphs

Authors: Jing Ma, Ruocheng Guo, Saumitra Mishra, Aidong Zhang, Jundong Li

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on both synthetic and real-world graphs validate the superiority of CLEAR over the state-of-the-art methods in different aspects. In this section, we evaluate our framework CLEAR with extensive experiments on both synthetic and real-world graphs.
Researcher Affiliation Collaboration Jing Ma University of Virginia Charlottesville, VA, USA jm3mr@virginia.edu Ruocheng Guo Bytedance AI Lab London, UK rguo.asu@gmail.com Saumitra Mishra J.P. Morgan AI Research London, UK saumitra.mishra@jpmorgan.com Aidong Zhang University of Virginia Charlottesville, VA, USA aidong@virginia.edu Jundong Li University of Virginia Charlottesville, VA, USA jundong@virginia.edu
Pseudocode No The paper describes the components and mechanism of CLEAR but does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code Yes The source code of our proposed framework is included in the supplemental material. It will be released after publication.
Open Datasets Yes We evaluate our method on three datasets, including a synthetic dataset and two datasets with realworld graphs. (1) Community. This dataset contains synthetic graphs generated by the Erdös-Rényi (E-R) model [24]. (2) Ogbg-molhiv. In this dataset, each graph stands for a molecule... (3) IMDB-M. This dataset contains movie collaboration networks from IMDB. In the ethics checklist, it states: 'The creators developed the adopted datasets based on either publicly available data or data with proper consent.'
Dataset Splits No The provided text indicates that data splits were specified in Appendix B ('Did you specify all the training details (e.g., data splits, hyperparameters, how they were chosen)? [Yes] See Section 4, Appendix B and C.'), but the main body of the paper does not explicitly detail the training/validation/test dataset splits with percentages or sample counts.
Hardware Specification No The paper states that 'total amount of compute and the type of resources used' are in Appendix B ('Did you include the total amount of compute and the type of resources used (e.g., type of GPUs, internal cluster, or cloud provider)? [Yes] See Appendix B.'). However, no specific hardware details (e.g., GPU models, CPU types) are provided in the main text.
Software Dependencies No The paper does not provide specific software dependencies (e.g., library names with version numbers like PyTorch 1.9 or Python 3.8) needed to replicate the experiment in the provided text.
Experiment Setup Yes We set the desired label Y for each graph as its flipped label (e.g., if Y = 0, then Y = 1). For each graph, we generate three counterfactuals for it (N CF = 3). Other setup details are in Appendix B. We vary α {0.01, 0.1, 1.0, 5.0, 10.0} and β {0.01, 0.1, 1.0, 10.0, 100}.