Classification dans les graphes hétérogènes basée sur une représentation latente des noeuds

Social networks are often composed of different types of nodes. Learning and performing inference on such heterogeneous networks is a recent task. We address the tasks of tagging of nodes in social networks, where the different types of nodes have to be labeled by different set of categories or tags. We propose a new transductive approach that automatically learns to project the different types of nodes onto a common latent space, this learned representation being then used for classifying the different elements. This framework exploits the idea that two nodes connected in a social network will tend to have a similar latent representation regardless of their type. This assumption allows us to learn correlations between the labels of nodes of different types, when state-of-the-art methods usually address each type of node separately. The model is tested on two datasets and shows good performance. MOTS-CLÉS : Classification dans les graphes, Graphes hétérogènes, Réseaux sociaux