Titre :	Community Detection with Deep Graph Neural Networks
Type de document :	texte imprimé
Auteurs :	Imad Eddine Kettaf, Auteur ; Abdelouahab Moussaoui, Directeur de thèse
Année de publication :	2022
Importance :	1 vol (79 f .)
Format :	29cm
Langues :	Français (fre)
Catégories :	Thèses & Mémoires:Informatique
Mots-clés :	Informatique
Index. décimale :	004 Informatique
Résumé :	n recent years, Deep Neural Networks (DNN) have achieved significant advances in image categorization and speech recognition. Using DNN for community identification, such as in social networks, remains a difficult challenge. Furthermore, the latest sophisticated techniques of executing community detection tasks employ Graph Neural Networks (GNN), a novel DNN methodology. In this thesis, we investigated the applicability of these new techniques for detecting communities by creating a new GNN-based learning model that combines two of the most recent state-of-the-art approaches in community detection, Graph Convolutional Networks (GCN) and Graph Attention Networks (GAT). Our suggested model is made up of four layers and performs semi-supervised node classification to discover communities. The first two layers are GCN layers, followed by a GAT layer, then ended with another GCN layer. We utilized four distinct datasets from the pytorch-geometric package for training and testing our model. These datasets are ”Cora,” ”Citeseer,” ”Pubmed,” and ”Ego-Facebook.” This architecutre of our model enabled it to outperform the other comparative techniques as well as certain state-of-the-art approaches.
Côte titre :	MAI/0607
En ligne :	https://drive.google.com/file/d/1bo_KtDR87-Ef9Z47Q7vUuWqfQnQmG0Q9/view?usp=share [...]
Format de la ressource électronique :	pdf

Community Detection with Deep Graph Neural Networks [texte imprimé] / Imad Eddine Kettaf, Auteur ; Abdelouahab Moussaoui, Directeur de thèse . - 2022 . - 1 vol (79 f .) ; 29cm.
Langues : Français (fre)

Catégories :	Thèses & Mémoires:Informatique
Mots-clés :	Informatique
Index. décimale :	004 Informatique
Résumé :	n recent years, Deep Neural Networks (DNN) have achieved significant advances in image categorization and speech recognition. Using DNN for community identification, such as in social networks, remains a difficult challenge. Furthermore, the latest sophisticated techniques of executing community detection tasks employ Graph Neural Networks (GNN), a novel DNN methodology. In this thesis, we investigated the applicability of these new techniques for detecting communities by creating a new GNN-based learning model that combines two of the most recent state-of-the-art approaches in community detection, Graph Convolutional Networks (GCN) and Graph Attention Networks (GAT). Our suggested model is made up of four layers and performs semi-supervised node classification to discover communities. The first two layers are GCN layers, followed by a GAT layer, then ended with another GCN layer. We utilized four distinct datasets from the pytorch-geometric package for training and testing our model. These datasets are ”Cora,” ”Citeseer,” ”Pubmed,” and ”Ego-Facebook.” This architecutre of our model enabled it to outperform the other comparative techniques as well as certain state-of-the-art approaches.
Côte titre :	MAI/0607
En ligne :	https://drive.google.com/file/d/1bo_KtDR87-Ef9Z47Q7vUuWqfQnQmG0Q9/view?usp=share [...]
Format de la ressource électronique :	pdf