MultiWalk: A Framework to Generate Node Embeddings Based on an Ensemble of Walk Methods

TL;DR

MultiWalk is a framework that combines multiple random walk-based methods to generate more effective graph node embeddings, improving classification performance over individual methods.

Contribution

It introduces an ensemble approach to graph embedding generation, leveraging multiple walk methods for enhanced accuracy.

Findings

Ensemble embeddings outperform individual methods in classification tasks.

Using two state-of-the-art walk methods improves embedding quality.

MultiWalk demonstrates the benefit of combining multiple embedding techniques.

Abstract

Graph embeddings are low dimensional representations of nodes, edges or whole graphs. Such representations allow for data in a network format to be used along with machine learning models for a variety of tasks (e.g., node classification), where using a similarity matrix would be impractical. In recent years, many methods for graph embedding generation have been created based on the idea of random walks. We propose MultiWalk, a framework that uses an ensemble of these methods to generate the embeddings. Our experiments show that the proposed framework, using an ensemble composed of two state-of-the-art methods, can generate embeddings that perform better in classification tasks than each method in isolation.