[Paper Review] Inferring Multiplex Diffusion Network via Multivariate Marked Hawkes Process

Peiyuan Sun, Jianxin Li+3 more|arXiv (Cornell University)|Aug 24, 2018

Diffusion and Search Dynamics19 references3 citations

TL;DR

This paper proposes the Multiplex Diffusion Model (MDM), a Bayesian generative model that infers multiplex social network structures by combining multivariate marked Hawkes processes with topic modeling. By leveraging event timestamps and content semantics, MDM jointly infers layered network topologies and influence strengths, outperforming baselines in network inference AUC across synthetic and real-world datasets.

ABSTRACT

Understanding the diffusion in social network is an important task. However, this task is challenging since (1) the network structure is usually hidden with only observations of events like "post" or "repost" associated with each node, and (2) the interactions between nodes encompass multiple distinct patterns which in turn affect the diffusion patterns. For instance, social interactions seldom develop on a single channel, and multiple relationships can bind pairs of people due to their various common interests. Most previous work considers only one of these two challenges which is apparently unrealistic. In this paper, we study the problem of \emph{inferring multiplex network} in social networks. We propose the Multiplex Diffusion Model (MDM) which incorporates the multivariate marked Hawkes process and topic model to infer the multiplex structure of social network. A MCMC based algorithm is developed to infer the latent multiplex structure and to estimate the node-related parameters. We evaluate our model based on both synthetic and real-world datasets. The results show that our model is more effective in terms of uncovering the multiplex network structure.

Motivation & Objective

• To address the challenge of inferring hidden, multiplex network structures in social networks where only event timestamps and content are observed.
• To model heterogeneous influence patterns across multiple interaction layers (e.g., politics, sports, lifestyle) that coexist between the same node pairs.
• To incorporate content semantics (via topic modeling) to distinguish between different diffusion channels in multiplex networks.
• To develop a scalable inference algorithm capable of handling intractable posterior distributions in complex, layered network structures.
• To improve the accuracy of network structure inference by explicitly modeling multiplexity and content-driven influence pathways.

Proposed method

• Formulates a multivariate marked Hawkes process to model the temporal dynamics of cascades across multiple layers of interaction.
• Integrates a mixed-membership topic model to assign content topics to each cascade event, enabling layer-specific influence modeling.
• Uses a Metropolis-within-Gibbs MCMC algorithm to jointly infer latent multiplex network structure, influence parameters, and topic assignments.
• Applies a slab-and-spike prior over edges to encourage sparsity and support multiplex structure discovery.
• Models each node’s influence as a function of its topic-specific transmission rates across layers.
• Leverages both event timestamps and textual content to disambiguate between multiple interaction channels between the same node pairs.

Experimental results

Research questions

• RQ1 Can a unified probabilistic model jointly infer multiplex network structures and influence parameters from timestamped, content-rich diffusion events?
• RQ2 How does incorporating topic information improve the accuracy of hidden network structure inference compared to models relying only on event timing?
• RQ3 To what extent can the model detect layered community structures based on content similarity and temporal diffusion patterns?
• RQ4 How does the model perform in comparison to state-of-the-art baselines in terms of AUC for network inference across varying network complexity?
• RQ5 What is the impact of increasing the number of layers on the model’s inference performance and scalability?

Key findings

• The MDM model consistently outperforms baseline models in network inference AUC across all three real-world datasets, with the best performance observed on the smallest dataset (dataset1).
• Performance degrades slightly with increasing number of layers, likely due to insufficient event density per layer for reliable training.
• The model achieves improved AUC on datasets with higher average events per edge, indicating better performance with richer data.
• Incorporating content semantics significantly enhances the model’s ability to distinguish between multiple interaction channels between the same node pairs.
• The model successfully detects community structures aligned with topic preferences (e.g., politics, sports, lifestyle) based on cascade content.
• The MCMC-based inference algorithm effectively explores the intractable posterior distribution, enabling joint estimation of network structure, influence strengths, and topic memberships.