Communications of the ACM
ML Model Could Remove Bias From Social Network Connections
Researchers at the Penn State College of Information Sciences and Technology have developed a novel framework which they say could remove bias from social network connections.
Credit: Apinan/Adobe Stock
A novel machine learning framework developed by Pennsylvania State University (Penn State) researchers can calculate sensitive attributes to help graph neural networks (GNNs) make fair recommendations.
The researchers trained the FairGNN model with user profiles on the Pokec Slovakian social network, and a dataset of about 400 National Basketball Association (NBA) players.
The Penn State researchers treated the region in which each Pokec user was from as the sensitive attribute, and tasked the classifier to predict their working field; for the NBA data, they identified players in the U.S. and those abroad, using location as the sensitive attribute for predicting whether each player's salary exceeded the median.
FairGNN upheld high-performance node classification using limited, user-supplied sensitive information while reducing bias.
Penn State's Suhang Wang said, "Our experiment shows that the classification performance doesn’t decrease. But in terms of fairness, we can make the model much more fair."
From Penn State News
View Full Article
Abstracts Copyright © 2021 SmithBucklin, Washington, DC, USA
No entries found