Inferring Diffusion Structures of Heterogeneous Network Cascade

2025-06-23

AI Summary

This paper addresses the critical problem of inferring diffusion structures in heterogeneous network cascades. We propose the Dual-Heterogeneous Directed Graph Model (DHDM), the first method to embed layer-specific structural constraints into a multilayer Granger causality framework. DHDM employs a provably convergent convex optimization algorithm, achieving statistically consistent and computationally efficient network reconstruction. Departing from conventional single-layer assumptions, it explicitly models both cross-layer heterogeneity and intra-layer causal dependencies, enabling interpretable identification of complex propagation pathways. In synthetic experiments, DHDM significantly outperforms state-of-the-art baselines in accurately recovering diverse realistic diffusion topologies. Empirical analysis on social science topic diffusion across U.S. universities reveals hierarchical inter-disciplinary information spillover mechanisms, uncovering previously unobserved causal structures. The method thus provides both theoretical foundations and practical tools for predictive modeling and targeted intervention in multilayer diffusion processes.

Technology Category

• Reasoning under Uncertainty: Causality
• Machine Learning: Causal Learning
• 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
• Social Networks and Social Media: Influence propagation, information diffusion, and the prediction on networks
• Web Mining and Content Analysis: Content-based information diffusion

Abstract

Network cascade refers to diffusion processes in which outcome changes within part of an interconnected population trigger a sequence of changes across the entire network. These cascades are governed by underlying diffusion networks, which are often latent. Inferring such networks is critical for understanding cascade pathways, uncovering Granger causality of interaction mechanisms among individuals, and enabling tasks such as forecasting or maximizing information propagation. In this project, we propose a novel double mixture directed graph model for inferring multi-layer diffusion networks from cascade data. The proposed model represents cascade pathways as a mixture of diffusion networks across different layers, effectively capturing the strong heterogeneity present in real-world cascades. Additionally, the model imposes layer-specific structural constraints, enabling diffusion networks at different layers to capture complementary cascading patterns at the population level. A key advantage of our model is its convex formulation, which allows us to establish both statistical and computational guarantees for the resulting diffusion network estimates. We conduct extensive simulation studies to demonstrate the model's performance in recovering diverse diffusion structures. Finally, we apply the proposed method to analyze cascades of research topics in the social sciences across U.S. universities, revealing the underlying diffusion networks of research topic propagation among institutions.

Problem

• Infer latent multi-layer diffusion networks from cascade data
• Capture heterogeneous cascading patterns across network layers
• Model research topic propagation among social science institutions

Innovation

• Double mixture directed graph model
• Layer-specific structural constraints
• Convex formulation guarantees