Hierarchical Mixed Neural Network for Joint Representation Learning of Social-Attribute Network

Most existing network representation learning (NRL) methods are designed for homogeneous network, which only consider topological properties of networks. However, in real-world networks, text or categorical attributes are usually associated with nodes, providing another description for networks in a different perspective. In this paper, we present a joint learning approach which learns the representations of nodes and attributes in the same low-dimensional vector space simultaneously. Particularly, we show that more discriminative node representations can be acquired by leveraging attribute features. The experiments conducted on three social-attribute network datasets demonstrate that our model outperforms several state-of-the-art baselines significantly for node classification task and network visualization task.