Deep Representation Learning on Large Attributed Graphs

Attributed graphs convey a rich set of information on relations, nodes, and attributes. However, the entire set of information cannot be easily captured by traditional Representation Learning methods usually adopted for creating low-dimension and meaningful embeddings of the graph. This research project aims to generate a representation that encodes both the relational structure of a graph and the node attributes that can be texts or images. The final goal is to obtain an unsupervised model based on promising Deep Learning architectures that is able to efficiently and effectively derive dense vector representations of nodes in attributed graphs.