Deep Identification of Propagation Trees
Zeeshan Memon, Chen Ling, Ruochen Kong, Vishwanath Seshagiri, Andreas Zufle, Liang Zhao
TL;DR
DIPT tackles the problem of identifying propagation trees from observed diffusion states, addressing unknown diffusion mechanisms, expansive tree spaces, and partial observations. It jointly learns a local node-influence model conditioned on node attributes and a latent seed-prior via variational inference, using an alternating optimization scheme to infer propagation trees without direct tree supervision. The method demonstrates strong propagation-tree identification and competitive source localization across five datasets, including real diffusion cascades and a large-scale simulated infectious-disease scenario, while also improving diffusion-state reconstruction and benefiting from partial supervision. Overall, DIPT provides a principled, data-driven approach to uncover transmission pathways and diffusion dynamics in complex networks, with practical implications for epidemiology and information diffusion analysis.
Abstract
Understanding propagation structures in graph diffusion processes, such as epidemic spread or misinformation diffusion, is a fundamental yet challenging problem. While existing methods primarily focus on source localization, they cannot reconstruct the underlying propagation trees i.e., "who infected whom", which are substantial for tracking the propagation pathways and investigate diffusion mechanisms. In this work, we propose Deep Identification of Propagation Trees (DIPT), a probabilistic framework that infers propagation trees from observed diffused states. DIPT models local influence strengths between nodes and leverages an alternating optimization strategy to jointly learn the diffusion mechanism and reconstruct the propagation structure. Extensive experiments on five real-world datasets demonstrate the effectiveness of DIPT in accurately reconstructing propagation trees.