Nowadays, many entities and the relationships between them create different types of networks such as social networks, biological networks, wireless networks and etc. Graphs are appropriate tools to model these networks. In a graph, nodes show individuals and edges show the relations between them. On the other hand, analyzing networks provides a good insight about the information hidden in them. Network analysis has different application such as node classification, node clustering, link prediction and etc. The main challenge of network analysis is to extract the most important
features of graph and to discriminate important and unimportant features. Specially, when the size of the network grows, the space cost and computational order are so challenging. Network embedding is a way to extract features for entities in graph in low dimension by representing it as a set of low dimensional vectors for every node and discard unimportant information in network. The main goal of graph embedding is to preserve graph structure, but nodes in the input graphs can also have attributes. For example, in a social network every member has age, gender, and other attributes (which are independent of the relationship between users), so it’s important to embed node attributes and relations at the same time in the embedding approach in a way that similar nodes have similar features in the new low-dimension space. Our purpose is to design a new approach for extracting features in networks that preserves structure and attribute proximity in the network with different types of nodes and edges.
Resource of image: https://www.slideshare.net/jleskovec/graph-representation-learning
People Involved: Mahsa Ghorbani, Mahdieh Soleymani, Hamid R. Rabiee