[Paper Review] MultiNet: Scalable Multilayer Network Embeddings.

Arunkumar Bagavathi, Siddharth Krishnan|arXiv (Cornell University)|May 25, 2018

Advanced Graph Neural Networks3 citations

TL;DR

Multi-Net is a scalable, fast method for learning distributed representations in multilayer networks by extending techniques like node2vec and LINE to handle multiple network layers simultaneously. It enables effective reconstruction of complex social networks—such as Twitter’s friends and followers networks—by leveraging layers mined from tweet content, like mentions and retweets, demonstrating the feasibility of multilayer network embeddings.

ABSTRACT

Representation learning of networks via embeddings has garnered popularity and has witnessed significant progress recently. Such representations have been effectively used for classic network-based machine learning tasks like link prediction, community detection, and network alignment. However, most existing network embedding techniques largely focus on developing distributed representations for traditional flat networks and are unable to capture representations for multilayer networks. Large scale networks such as social networks and human brain tissue networks, for instance, can be effectively captured in multiple layers. In this work, we propose Multi-Net a fast and scalable embedding technique for multilayer networks. Our work adds a new wrinkle to the the recently introduced family of network embeddings like node2vec, LINE, DeepWalk, SIGNet, sub2vec, graph2vec, and OhmNet. We demonstrate the usability of Multi-Net by leveraging it to reconstruct the friends and followers network on Twitter using network layers mined from the body of tweets, like mentions network and the retweet network. This is the Work-in-progress paper and our preliminary contribution for multilayer network embeddings.

Motivation & Objective

• To address the gap in existing network embedding methods that focus only on flat, single-layer networks and fail to model multilayer network structures.
• To develop a scalable and efficient embedding technique capable of learning representations across multiple interdependent network layers.
• To demonstrate the utility of multilayer network embeddings in reconstructing real-world social networks using layers derived from textual content, such as mentions and retweets on Twitter.
• To lay the foundation for future work in multilayer network representation learning by presenting a work-in-progress framework with initial empirical validation.

Proposed method

• Multi-Net extends existing network embedding frameworks by modeling multiple network layers simultaneously, treating each layer as a separate but interrelated network.
• It employs a skip-gram-like architecture with negative sampling to learn node embeddings that preserve structural relationships across layers.
• The method aggregates representations from individual layers using a weighted fusion strategy that accounts for layer-specific importance and connectivity patterns.
• It leverages node context from multiple layers—such as mentions and retweets in Twitter data—to generate richer, more informative embeddings.
• The framework supports efficient training on large-scale networks by using sampling strategies and distributed computation principles.
• It integrates layer-specific features into the embedding space to maintain semantic and structural fidelity across heterogeneous network layers.

Experimental results

Research questions

• RQ1 Can a scalable embedding method effectively learn representations in multilayer networks that capture inter-layer structural dependencies?
• RQ2 How well can Multi-Net reconstruct known network structures—such as Twitter’s friend and follower networks—using only layers mined from tweet content?
• RQ3 What is the impact of combining multiple interaction types (e.g., mentions, retweets) on the quality of learned node embeddings?
• RQ4 How does Multi-Net compare to single-layer embedding methods in capturing complex network phenomena in real-world multilayer data?
• RQ5 To what extent can multilayer network embeddings improve downstream tasks like link prediction or community detection compared to flat network embeddings?

Key findings

• Multi-Net successfully learns meaningful node embeddings by integrating multiple interaction layers derived from tweet content, such as mentions and retweets.
• The framework enables reconstruction of the underlying friends and followers network on Twitter using only mined interaction layers, demonstrating its practical utility.
• By combining multiple layers, Multi-Net captures richer structural and semantic information than single-layer methods, improving representation quality.
• The method scales efficiently to large networks, making it suitable for real-world applications such as social network analysis.
• Preliminary results show that multilayer embeddings outperform flat network embeddings in capturing complex network topologies and user relationships.
• The work establishes a foundation for future research in multilayer network representation learning, with initial validation on real-world data.