Graph Enhanced Feature Engineering for Privacy Preserving Recommendation Systems

Abstract

This paper presents a solution for RecSys Challenge 2023 which leverages 1) a novel feature classification method to categorize anonymous features into distinct groups and apply enhanced feature engineering and 2) data modeling as bipartite and similarity graphs where supervised and unsupervised learning can be applied to reveal underlying information in the form of new features and improve prediction accuracy of an ensemble of classifiers. Our team's name is LearningFE, we rank 2nd on the leader board (score=5.892977).