Journal
INFORMATION SCIENCES
Volume 495, Issue -, Pages 37-51Publisher
ELSEVIER SCIENCE INC
DOI: 10.1016/j.ins.2019.05.001
Keywords
Network embedding; Node similarity; Deep learning; Autoencoders
Categories
Funding
- National key research and development program of China [2017YFB0802200]
- Key research and development program of Shaanxi Province [2018ZDXM-GY-045]
- Fundamental Research Funds for Central Universities
- Innovation Fund of Xidian University
Ask authors/readers for more resources
Networks such as social networks, computer networks, and citation networks are ubiquitous in the real world. Aiming to learn distributed representations for nodes in a network, network embedding has attracted widespread attention from quite a few researchers due to its critical application in complex network analysis. However, the majority of existing algorithms merely facilitate the local pairwise proximity or implicitly characterize the neighborhood similarity in a network, which are not expressive enough to characterize the diversity of node similarities. In this paper, we present an innovative network embedding framework, Sim2vec, which can encode more comprehensive node similarities among different nodes in a network into unified latent spaces based on deep neural networks. Specifically, we incorporate the adjacency similarity, the accessibility similarity and the neighborhood similarity between two nodes, and introduce their respective measures. To capture more integral structural information in a network, these node similarities are encoded into low-dimensional and dense vector spaces by jointly optimizing the carefully designed objective function. Extensive experiments conducted on several networks demonstrate that the proposed approach is effective and it outperforms the existing state-of-the-art approaches on a variety of tasks, including node classification, link prediction, node visualization and node clustering. (C) 2019 Elsevier Inc. All rights reserved.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available