Toward Edge-Centric Network Embeddings

TL;DR

This paper introduces edge-centric network embeddings using line graphs and an edge weighting mechanism, enabling improved link prediction by embedding and aggregating edges along paths between node pairs.

Contribution

It proposes ECNE, an edge embedding method based on line graphs, and ECNE-LP, a link prediction framework that leverages edge embeddings along paths.

Findings

ECNE outperforms node-centric embeddings in certain tasks.

ECNE-LP improves link prediction accuracy.

Edge embeddings capture dynamic graph information effectively.

Abstract

Existing network embedding approaches tackle the problem of learning low-dimensional node representations. However, networks can also be seen in the light of edges interlinking pairs of nodes. The broad goal of this paper is to introduce edge-centric network embeddings. We present an approach called ECNE, which instead of computing node embeddings directly, computes edge embeddings by relying on the notion of line graph coupled with an edge weighting mechanism to preserve the dynamic of the original graph in the line graph. We also present a link prediction framework called ECNE-LP, which given a target link (u,v) first collects paths between nodes u and v, then directly embeds the edges in these paths, and finally aggregates them toward predicting the existence of a link. We show that both ECNE and ECNE-LP bring benefit wrt the state-of-the-art.