This paper introduces TEMPUS, a novel method for community detection in temporal networks where the number of communities evolves over time. TEMPUS employs a distance-dependent Chinese restaurant process and a temporal smoothness constraint within a degree-corrected stochastic block model to adaptively infer community structures without pre-specifying the number of communities. The method includes an efficient variational inference algorithm and is evaluated on synthetic and real-world datasets.
Key findings
TEMPUS can detect communities in temporal networks with varying community counts.
The method integrates a distance-dependent Chinese restaurant process and temporal smoothness constraint.
An efficient variational inference algorithm scales to networks with millions of edges.
Comprehensive experimental validation on synthetic and real-world datasets.
Limitations & open questions
The paper is a research proposal and may not cover all potential limitations of the TEMPUS framework.
The proposed method's performance in extremely large-scale networks or with highly dynamic community structures is not fully explored.