GNE: Generic Heterogeneous Information Network Embedding

As an effective approach to solve graph mining problems, network embedding aims to learn low-dimensional latent representation of nodes in a network. We develop a representation learning method called GNE for generic heterogeneous information networks to learn the vertex representations for generic HINs. Greatly different from previous works, our model consists two components. First, GNE assigns the probability of each random walk step according to vertex centrality, weight of relations and structural similarity for neighbors on premise of performing a biased self-adaptive random walk generator. Second, to learn more desirable representations for generic HINs, we then design an advanced joint optimization framework by accounting for both the explicit (1st-order) relations and implicit (higher-order) relations.