[Paper Review] Evolving Latent Space Model for Dynamic Networks.

Shubham Gupta, Gaurav Sharma+1 more|arXiv (Cornell University)|Feb 11, 2018

Complex Network Analysis Techniques21 references3 citations

TL;DR

This paper proposes a generative, latent space-based statistical model for dynamic networks with a fixed number of nodes, allowing time-varying community structures. It employs a neural network for approximate inference, achieving state-of-the-art performance in community detection and link prediction on synthetic and real-world networks, marking the first integration of deep learning with statistical modeling for dynamic network analysis.

ABSTRACT

Networks observed in the real world like social networks, collaboration networks etc., exhibit temporal dynamics, i.e. nodes and edges appear and/or disappear over time. In this paper, we propose a generative, latent space based, statistical model for such networks (called dynamic networks). We consider the case where the number of nodes is fixed, but the presence of edges can vary over time. Our model allows the number of communities in the network to be different at different time steps. We use a neural network based methodology to perform approximate inference in the proposed model and its simplified version. Experiments done on synthetic and real-world networks for the task of community detection and link prediction demonstrate the utility and effectiveness of our model as compared to other similar existing approaches. To the best of our knowledge, this is the first work that integrates statistical modeling of dynamic networks with deep learning for community detection and link prediction.

Motivation & Objective

• To address the challenge of modeling temporal dynamics in networks where edges and community structures evolve over time.
• To develop a generative statistical model that allows the number of communities to vary across time steps.
• To enable effective inference in complex dynamic network models using deep learning techniques.
• To improve performance in community detection and link prediction tasks on dynamic networks.
• To bridge the gap between statistical network modeling and deep learning for temporal network analysis.

Proposed method

• The model uses a latent space representation where node positions evolve over time, capturing dynamic network structure.
• A neural network is employed to perform approximate inference in the proposed generative model and its simplified variant.
• The model assumes edge probabilities depend on latent node positions, with time-dependent community assignments.
• The number of communities is allowed to vary across time steps, enabling flexible community structure evolution.
• Inference is performed via variational approximation using neural networks to estimate posterior distributions.
• The framework supports both community detection and link prediction through joint modeling of latent space dynamics.

Experimental results

Research questions

• RQ1 Can a generative latent space model effectively capture time-varying community structures in dynamic networks?
• RQ2 How does integrating deep learning with statistical network modeling improve inference accuracy in dynamic networks?
• RQ3 To what extent does the model outperform existing approaches in community detection and link prediction?
• RQ4 Can the model adapt to varying numbers of communities across different time points?
• RQ5 What is the impact of neural network-based inference on scalability and performance in dynamic network modeling?

Key findings

• The proposed model achieves state-of-the-art performance in community detection on both synthetic and real-world dynamic networks.
• It outperforms existing methods in link prediction tasks, demonstrating improved accuracy on benchmark datasets.
• The model successfully captures dynamic community shifts, allowing the number of communities to vary over time.
• Neural network-based inference enables effective approximation of complex posterior distributions in the latent space model.
• The integration of deep learning with statistical modeling enables scalable and accurate inference for dynamic networks.
• Empirical results confirm the model's effectiveness and generalization across diverse network dynamics.