Relational learning via latent social dimensions

This work proposes to extract latent social dimensions based on network information, and then utilize them as features for discriminative learning, and outperforms representative relational learning methods based on collective inference, especially when few labeled data are available.