[論文レビュー] MultiVERSE: a multiplex and multiplex-heterogeneous network embedding approach

Léo Pio‐Lopez, Alberto Valdeolivas+3 more|arXiv (Cornell University)|Aug 23, 2020

Advanced Graph Neural Networks参考文献 148被引用数 50

ひとこと要約

tldr: MultiVERSE は VERSE フレームワークを Random Walks with Restart on Multiplex (RWR-M) および Multiplex-Heterogeneous (RWR-MH) で拡張し、 multiplex および multiplex-heterogeneous ネットワークからノード埋め込みを学習し、リンク予測およびネットワーク再構築性能の向上を示す。

ABSTRACT

Network embedding approaches are gaining momentum to analyse a large variety of networks. Indeed, these approaches have demonstrated their efficiency for tasks such as community detection, node classification, and link prediction. However, very few network embedding methods have been specifically designed to handle multiplex networks, i.e. networks composed of different layers sharing the same set of nodes but having different types of edges. Moreover, to our knowledge, existing approaches cannot embed multiple nodes from multiplex-heterogeneous networks, i.e. networks composed of several layers containing both different types of nodes and edges. In this study, we propose MultiVERSE, an extension of the VERSE method with Random Walks with Restart on Multiplex (RWR-M) and Multiplex-Heterogeneous (RWR-MH) networks. MultiVERSE is a fast and scalable method to learn node embeddings from multiplex and multiplex-heterogeneous networks. We evaluate MultiVERSE on several biological and social networks and demonstrate its efficiency. MultiVERSE indeed outperforms most of the other methods in the tasks of link prediction and network reconstruction for multiplex network embedding, and is also efficient in the task of link prediction for multiplex-heterogeneous network embedding. Finally, we apply MultiVERSE to study rare disease-gene associations using link prediction and clustering. MultiVERSE is freely available on github at https://github.com/Lpiol/MultiVERSE.

研究の動機と目的

• Motivate the need for embedding methods tailored to multiplex and multiplex-heterogeneous networks.
• Extend VERSE with RWR-M and RWR-MH to learn node embeddings across multiple layers and node types.
• Evaluate performance on diverse biological and social networks for link prediction and network reconstruction.
• Demonstrate application to rare disease-gene associations through link prediction and clustering.
• Provide an open-source implementation for the research community.

提案手法

• Adopt VERSE as the general embedding framework and minimize KL-divergence between graph-based and embedding-based similarity distributions using Noise Contrastive Estimation (NCE).
• Define similarity distributions sim_G via Random Walk with Restart on Multiplex (RWR-M) and on Multiplex-Heterogeneous networks (RWR-MH).
• Represent multiplex graphs with L layers sharing nodes, and construct corresponding multiplex- and bipartite-connected transition structures for RWR-M and RWR-MH.
• Run RWR-M and RWR-MH to obtain node-to-node similarity distributions suitable as VERSE input, then train embeddings with stochastic updates and NCE biases.
• In MultiVERSE, compute embeddings by sampling node pairs from similarity distributions, applying embedding updates via a binary classifier objective with NCE (bias terms included).
• Provide a public GitHub release for reproducibility.

実験結果

リサーチクエスチョン

• RQ1 Can MultiVERSE effectively embed nodes in multiplex networks without merging layers, preserving layer-specific topology?
• RQ2 Can MultiVERSE extend to multiplex-heterogeneous networks to embed multiple node types?
• RQ3 How does MultiVERSE perform on link prediction and network reconstruction tasks compared to existing monoplex and multiplex embedding methods?
• RQ4 Is MultiVERSE scalable to large multiplex and multiplex-heterogeneous networks and adaptable to biological and social datasets?
• RQ5 Can MultiVERSE enable downstream biological insights such as gene-disease association analyses through link prediction and clustering?

主な発見

• MultiVERSE outperforms state-of-the-art methods on link prediction and network reconstruction for multiplex networks.
• MultiVERSE provides the first multiplex-heterogeneous embedding method that handles embedding across different node types.
• The evaluation across seven multiplex datasets from biological, neuronal, co-authorship and social domains demonstrates efficiency and scalability.
• The framework supports applications such as studying rare disease–gene associations via link prediction and clustering.
• The approach is implemented in publicly available software (GitHub: Lpiol/MultiVERSE).