Repository to demonstrate codes, instructions, dependencies (i.e., package requirements), and data, of the paper titled "A Deep Latent Factor Model for\Fairness-Utility Trade-off in Graph Clustering" accepted for IEEE BigData 2025.

Our code also provides comparisons to the spectral-based clustering models-- including vanilla SC, FairSC and scalable FairSC-models--, as well as GNN-based and NMF-based baselines implemented in Python and utilizes the respective baseline implementations automatically. Start by installing the requirements.

In order to run the SBM experiments, redirect to the experiments directory and run the "sbm.py" script using python3. It itertaes over n=2000, n=5000, and, n=10000 number of nodes for k=5 clusters and h=2 groups as in our experiments and compared our model DFNMF with the baselines.

For different parameter setups, please change these values directly in the code snippet "sbm.py" and run.

We have provided the cleaned, pre-processed, ready-to-use real datasets used in the DFNMF\data directory. For each dataset, the data_loader function loads the corresponding dataset based on the input argument in the main function. In order to run the experiment for either of the real datasets, redirect to the experiments directory and run the "experiments.py" script. As also commented in the code, you need to select the dataset to execute analysis:

Please input the dataset ID as an ordinal number from $1-7$ in the configuration data structure in the main function to indicate the name of the dataset in the above order.

The code will automatically load the data using the proper data_loader and run a grid-search for DFNMF with a range of $\lambda \in [0.001,\cdots,1000]$, compared to the baselines for $k \in [2,10]$ clusters.