Multi-view Dynamic Heterogeneous Information Network Embedding

Abstract Most existing heterogeneous information network (HIN) embedding methods focus on static environments while neglecting the evolving characteristic of real-world networks. Although several dynamic have been proposed, they are merely designed for homogeneous networks and cannot be directly applied in environments. To tackle above challenges, we propose a novel framework incorporating temporal into HIN embedding, named multi-view (MDHNE), which can efficiently preserve evolution patterns implicit relationships from different views updating node vectors over time. We first transform to series corresponding views. Then our proposed MDHNE applies recurrent neural (RNN) incorporate pattern complex structure semantic between nodes latent spaces, thus multiple learned updated when evolves Moreover, come up with an attention-based fusion mechanism, automatically infer weights by minimizing objective function specific mining tasks. Extensive experiments clearly demonstrate that model outperforms state-of-the-art baselines three datasets