erasmopurif/FairUP · Apply for community grant: Academic project

This space is the official implementation of "FairUP: a Framework for Fairness Analysis of Graph Neural Network-Based User Profiling Models", co-authored by Mohamed Abdelrazed, Erasmo Purificato, Ludovico Boratto and Ernesto W. De Luca.
The paper has been accepted for publication at SIGIR 2023 conference.

FairUP is a standardised framework that empowers researchers and practitioners to simultaneously analyse state-of-the-art Graph Neural Network-based models for user profiling task, in terms of classification performance and fairness metrics scores.
The framework presents several components, which allow end-users to:

compute the fairness of the input dataset by means of a pre-processing fairness metric, i.e. disparate impact;
mitigate the unfairness of the dataset, if needed, by applying different debiasing methods, i.e. sampling, reweighting and disparate impact remover;
train one or more GNN models, specifying the parameters for each of them;
evaluate post-hoc fairness by exploiting four metrics, i.e. statistical parity, equal opportunity, overall accuracy equality, treatment equality.

For the best performance, the framework needs to run on a GPU, but due to cost constraints, we can't afford to pay for a GPU running every day. Thus, we are applying for a community GPU grant mostly to allow researchers and practitioners in the fields of Algorithmic Fairness and Graph Neural Networks to use FairUP in their experiments at its best at any time.