Using Network Embeddings for Improving Network Alignment

TL;DR

This paper proposes a framework that leverages network embeddings to enhance local network alignment algorithms, addressing biases in seed node selection and improving structural similarity detection.

Contribution

It introduces a novel framework combining network embeddings with local network alignment, advancing the accuracy of structural similarity detection.

Findings

Network embeddings improve alignment quality.

Embedding-based seed selection reduces bias.

Enhanced structural similarity detection.

Abstract

Network (or Graph) Alignment Algorithms aims to reveal structural similarities among graphs. In particular Local Network Alignment Algorithms (LNAs) finds local regions of similarity among two or more networks. Such algorithms are in general based on a set of seed nodes that are used to grow an alignment. Almost all LNAs algorithms use as seed nodes a set of vertices based on context information (e.g. a set of biologically related in biological network alignment) and this may cause a bias or a data-circularity problem. More recently, we demonstrated that the use of topological information in the choice of seed nodes may improve the quality of the alignments. We used some common approaches based on global alignment algorithms for capturing topological similarity among nodes. In parallel, it has been demonstrated that the use of network embedding methods (or representation learning), may capture the structural similarity among nodes better than other methods. Therefore we propose to use network embeddings to learn structural similarity among nodes and to use such similarity to improve LNA extendings our previous algorithms. We define a framework for LNA.