École Nationale Supérieure de
Techniques Avancées

Abstract

This report presents a study conducted during an internship that explores the application of Graph Neural Networks (GNNs) for binary link prediction within graphs. GNNs, constituting a dynamic branch of Artificial Intelligence (AI), stand out for their ability to model intricate relationships present in data in the form of graphs. Their application domain spans across various fields, including social networks, transportation systems, and power grids. This study took place within the context of a Kaggle competition associated with the "Conference on Data Science and Advanced Analytics" (DSAA), where the challenge revolved around predicting links within a given graph. This study specifically focuses on the implementation of a particular GNN architecture: SEAL [2]. The results obtained through this implementation showcase the model's efficiency with a prediction score of 0.98, while underscoring the pivotal significance of hyperparameter optimization in the process.