Clustering in Networks with Time-varying Nodal Attributes

2025-11-06

AI Summary

This paper addresses clustering on dynamic graphs whose node attributes evolve as time series, tackling the limitation of existing methods that ignore temporal evolution patterns in attributes. We propose a joint representation learning framework that simultaneously models graph structure and temporal dynamics. Methodologically, we design a collaborative mechanism integrating graph-regularized priors with a temporal decoder, incorporate a graph-fused LASSO constraint to enforce sparsity and interpretability, and perform Bayesian inference via Langevin dynamics. Model optimization employs maximum approximate likelihood estimation coupled with the alternating direction method of multipliers (ADMM). Extensive experiments on synthetic graphs (stochastic block models, grid graphs) and real-world datasets (temperature networks, text evolution networks) demonstrate that our approach significantly outperforms state-of-the-art baselines in clustering accuracy, while enhancing robustness and interpretability for node clustering in time-varying attributed networks.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Reasoning under Uncertainty: Graphical Models
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community

Application Category

• Graph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated Web-related graphs
• Web Mining and Content Analysis: Models for Web evolution
• Semantics and Knowledge: Methods, algorithms and applications for the development of semantic models, knowledge graphs and other forms of structured data models with machine-interpretable semantics

Abstract

This manuscript studies nodal clustering in graphs having a time series at each node. The framework includes priors for low-dimensional representations and a decoder that bridges the latent representations and time series. The structural and temporal patterns are fused into representations that facilitate clustering, addressing the limitation that the evolution of nodal attributes is often overlooked. Parameters are learned via maximum approximate likelihood, with a graph-fused LASSO regularization imposed on prior parameters. The optimization problem is solved via alternating direction method of multipliers; Langevin dynamics are employed for posterior inference. Simulation studies on block and grid graphs with autoregressive dynamics, and applications to California county temperatures and a book word co-occurrence network demonstrate the effectiveness of the proposed method.

Problem

• Clustering nodes in networks with time-varying nodal attributes
• Fusing structural and temporal patterns into latent representations
• Addressing overlooked evolution of nodal attributes in clustering

Innovation

• Fuses structural and temporal patterns into nodal representations
• Uses graph-fused LASSO regularization for prior parameters
• Employs alternating direction method for optimization