The content discusses the development of a dynamic edge partition model for temporal relational learning, focusing on capturing the evolution of vertices' memberships over time. The proposed model utilizes Dirichlet Markov chains and hierarchical beta-gamma priors to automatically infer latent communities and enable scalable inference through stochastic gradient MCMC algorithms. Experimental results demonstrate the accuracy and efficiency of the novel methods on various real-world datasets, showcasing superior performance in link prediction tasks compared to baseline models.
The content highlights the challenges in applying probabilistic dynamic network models to handle large graph-structured data efficiently. It introduces a new framework that incorporates side information and infers tree-structured latent community hierarchies. Future research directions include exploring privacy-preserving learning methods and incorporating advanced sampling techniques for dynamic network modeling.
To Another Language
from source content
arxiv.org
Key Insights Distilled From
by Sikun Yang,H... at arxiv.org 03-04-2024
https://arxiv.org/pdf/2403.00044.pdfDeeper Inquiries