PRRE: Personalized Relation Ranking Embedding for Attributed Networks

摘要: Attributed network embedding focuses on learning low-dimensional latent representations of nodes which can well preserve the original topological and node attributed proximity at same time. Existing works usually assume that with similar topology or attributes should also be close in space. This assumption ignores phenomenon partial correlation between similarities i.e. may dissimilar their vice versa. Partial two information sources considered especially when there exist fraudulent edges (i.e., from one source is vague) unbalanced data distributions (i.e, structure similarity attribute have different distributions). However, it very challenging to consider due heterogeneity these sources. In this paper, we take into account propose Personalized Relation Ranking Embedding (PRRE) method for networks capable exploiting attributes. The proposed PRRE model utilizes thresholds define relations employs Expectation-Maximization (EM) algorithm learn as other parameters. Extensive experiments results multiple real-world datasets show significantly outperforms state-of-the-art methods terms various evaluation metrics.