Adversarial enhanced representation for link prediction in multi-layer networks

Ruohan Yang, Junyang Chen, Huan Wang, Mengzhu Wang, Ziwen Cui, Victor C. M. Leung, Di Wang

Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: Link prediction; Multi-layer networks; Supervised representation learning; Adversarial training; Adaptive fusion

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

Abstract: Multi-layer networks are widely utilized in various applications, including social networks, biological networks, and Internet typologies. In these networks, link prediction is a longstanding issue that predicts missing links based on the observed structures across all layers, thereby assisting in tasks such as network recovery and drug-target prediction. However, existing link prediction methods tend to learn nontransferable intra-layer representations that cannot generalize well to other layers, which results in inefficient utilization of the structural correlations between layers in multi-layer networks. To address this, we propose a novel graph embedding method called Adversarial Enhanced Representation (AER) for link prediction in multi-layer networks. AER comprises three modules: a representation generator, a layer discriminator, and a link predictor. The representation generator is designed to learn and fuse the links’ inter-layer and intra-layer representations. Also, the layer discriminator aims to identify the layer sources of learned inter-layer representations. During a minimax two-player game, the representation generator attempts to learn inter-layer transferable representations to deceive the layer discriminator. In order not to be deceived, the layer discriminator attempts to accurately distinguish the layer sources of learned inter-layer representations. Finally, the link predictor works in collaboration with the representation generator to predict whether a link is a missing link based on the adaptive fusion between inter-layer transferable and intra-layer representations. To validate the effectiveness of our proposed method, we conduct extensive experiments on real-world datasets. The experimental results demonstrate that AER outperforms state-of-the-art methods in link prediction performance for multi-layer networks.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 3189