A repository for implementing and testing code from the paper FairEGM: Fair Link Prediction and Reccomendation via Emulated Graph Modification.
Sean Current, Yuntian He, Saket Gurukar, and Srinivasan Parthasarathy. "FairEGM: Fair Link Prediction and Reccomendation via Emulated Graph Modification." In ACM Conference on Equity and Access in Algorithms, Mechanisms, and Optimization. 2022.
FairEGM uses Tensorflow >2.5.0 and Scikit-Learn >1.0.0.
All datasets are stored in ./data. To run FairEGM experiments, use:
python main.py --dataset cora -d 32 -d2 16 -lr 0.0001 -e 300 -r 20
Where --dataset can be used to change the experimental dataset (must be one of citeseer, cora, facebook, or pubmed). The parameters -d and -d2 can be used to change the dimensions of the first and second GCN layer, respectively. The learning rate and number of training epochs can be changed with -lr and -e. Finally, -r specifies the number of random runs. For a single run, set -r to 1.
Code for FairEGM models are in the ./layers and ./models directories. The GFO, CFO, and FEW approaches are implemented as tf.keras.layers.Layer subclasses. These must be used in tandem with a FairModel object from ./models/fair_model.py, which stacks the GFO, CFO, and FEW convolutions with a learning layer and secondary GCN layer.