[Paper Review] A holistic approach for predicting links in coevolving multiplex networks

Alireza Hajibagheri, Gita Sukthankar+1 more|arXiv (Cornell University)|Aug 18, 2016

Complex Network Analysis Techniques27 references6 citations

TL;DR

This paper proposes MLP, a holistic framework for link prediction in coevolving multiplex networks that leverages inter-layer dependencies by learning link existence likelihoods from non-target layers to reweight single-layer prediction scores. By combining topological metrics via rank aggregation and reweighting based on cross-layer patterns, MLP outperforms factored approaches and fusion methods in predicting missing links across dynamic, multi-relational networks.

ABSTRACT

Networks extracted from social media platforms frequently include multiple types of links that dynamically change over time; these links can be used to represent dyadic interactions such as economic transactions, communications, and shared activities. Organizing this data into a dynamic multiplex network, where each layer is composed of a single edge type linking the same underlying vertices, can reveal interesting cross-layer interaction patterns. In coevolving networks, links in one layer result in an increased probability of other types of links forming between the same node pair. Hence we believe that a holistic approach in which all the layers are simultaneously considered can outperform a factored approach in which link prediction is performed separately in each layer. This paper introduces a comprehensive framework, MLP (Multiplex Link Prediction), in which link existence likelihoods for the target layer are learned from the other network layers. These likelihoods are used to reweight the output of a single layer link prediction method that uses rank aggregation to combine a set of topological metrics. Our experiments show that our reweighting procedure outperforms other methods for fusing information across network layers.

Motivation & Objective

• To address the limitation of factored link prediction methods that analyze each network layer in isolation, ignoring dynamic coevolution between link types.
• To model how the presence of links in one layer increases the likelihood of links forming in other layers, capturing cross-layer interaction patterns.
• To develop a unified framework that simultaneously leverages information from all layers to improve link prediction accuracy.
• To demonstrate that reweighting single-layer predictions using inter-layer likelihoods enhances performance over standard fusion techniques.

Proposed method

• The framework constructs a multiplex network where each layer represents a distinct type of dyadic interaction (e.g., communication, transactions) on the same set of nodes.
• For each target layer, the model learns the likelihood of link existence based on the presence of links in other layers using a probabilistic or statistical estimation method.
• A single-layer link prediction method is applied to each layer using a set of topological metrics, which are then ranked and aggregated via rank aggregation to produce a consensus prediction score.
• The aggregated scores are reweighted using the inter-layer link existence likelihoods, giving higher influence to predictions that are more likely to occur given the coevolving structure.
• The final prediction scores are used to rank node pairs, with higher scores indicating greater likelihood of missing links.

Experimental results

Research questions

• RQ1 Can modeling inter-layer dependencies in coevolving multiplex networks improve link prediction performance compared to treating layers independently?
• RQ2 How effective is reweighting single-layer predictions using inter-layer link existence likelihoods in enhancing prediction accuracy?
• RQ3 Does a holistic framework that considers all layers simultaneously outperform factored approaches that predict per-layer independently?
• RQ4 How do different topological metrics contribute to prediction when combined through rank aggregation and reweighting?

Key findings

• The MLP framework significantly outperforms standard factored link prediction methods that analyze each layer in isolation.
• The reweighting mechanism based on inter-layer link existence likelihoods improves prediction performance compared to baseline fusion strategies.
• Rank aggregation of topological metrics combined with inter-layer reweighting yields more robust and accurate predictions than individual metric performance.
• The holistic approach captures coevolutionary patterns that are missed by per-layer methods, especially in dynamic, multi-relational network settings.