Relation Representation Learning via Signed Graph Mutual Information Maximization for Trust Prediction

Trust prediction is essential to enhancing reliability and reducing risk from the unreliable node, especially for online applications in open network environments. An fact trust measure relation of both interacting entities accurately. However, most existing methods infer between usually rely on modeling similarity nodes a graph ignore semantic influence negative links (e.g., distrust relation). In this paper, we proposed representation learning via signed mutual information maximization (called SGMIM). SGMIM, incorporate translation model positive point-wise enhance representations adopt Mutual Information Maximization align entity spaces. Moreover, further develop sign making accurate predictions. We conduct link networks based learned representation. Extensive experimental results four real-world datasets task show that SGMIM significantly outperforms state-of-the-art baseline methods.