Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CausalMamba: Interpretable State Space Modeling for Temporal Rumor Causality

Xiaotong Zhan, Xi Cheng

TL;DR

CausalMamba tackles the problem of rumor detection with the need for interpretable, causally informed insights into how misinformation propagates. It combines a Mamba-based sequence encoder, a GCN for propagation structure, and a differentiable causal discovery module inspired by NOTEARS to jointly learn veracity and causal graphs. The approach achieves competitive classification performance while enabling counterfactual interventions and identification of influential nodes, offering actionable guidance for moderation. The work highlights the potential of integrating content, structure, and causality to produce explainable rumor diffusion models with practical impact for misinformation mitigation.

Abstract

Rumor detection on social media remains a challenging task due to the complex propagation dynamics and the limited interpretability of existing models. While recent neural architectures capture content and structural features, they often fail to reveal the underlying causal mechanisms of misinformation spread. We propose CausalMamba, a novel framework that integrates Mamba-based sequence modeling, graph convolutional networks (GCNs), and differentiable causal discovery via NOTEARS. CausalMamba learns joint representations of temporal tweet sequences and reply structures, while uncovering latent causal graphs to identify influential nodes within each propagation chain. Experiments on the Twitter15 dataset show that our model achieves competitive classification performance compared to strong baselines, and uniquely enables counterfactual intervention analysis. Qualitative results demonstrate that removing top-ranked causal nodes significantly alters graph connectivity, offering interpretable insights into rumor dynamics. Our framework provides a unified approach for rumor classification and influence analysis, paving the way for more explainable and actionable misinformation detection systems.

CausalMamba: Interpretable State Space Modeling for Temporal Rumor Causality

TL;DR

CausalMamba tackles the problem of rumor detection with the need for interpretable, causally informed insights into how misinformation propagates. It combines a Mamba-based sequence encoder, a GCN for propagation structure, and a differentiable causal discovery module inspired by NOTEARS to jointly learn veracity and causal graphs. The approach achieves competitive classification performance while enabling counterfactual interventions and identification of influential nodes, offering actionable guidance for moderation. The work highlights the potential of integrating content, structure, and causality to produce explainable rumor diffusion models with practical impact for misinformation mitigation.

Abstract

Rumor detection on social media remains a challenging task due to the complex propagation dynamics and the limited interpretability of existing models. While recent neural architectures capture content and structural features, they often fail to reveal the underlying causal mechanisms of misinformation spread. We propose CausalMamba, a novel framework that integrates Mamba-based sequence modeling, graph convolutional networks (GCNs), and differentiable causal discovery via NOTEARS. CausalMamba learns joint representations of temporal tweet sequences and reply structures, while uncovering latent causal graphs to identify influential nodes within each propagation chain. Experiments on the Twitter15 dataset show that our model achieves competitive classification performance compared to strong baselines, and uniquely enables counterfactual intervention analysis. Qualitative results demonstrate that removing top-ranked causal nodes significantly alters graph connectivity, offering interpretable insights into rumor dynamics. Our framework provides a unified approach for rumor classification and influence analysis, paving the way for more explainable and actionable misinformation detection systems.

Paper Structure

This paper contains 42 sections, 9 equations, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Structured Sequence Modeling for Long-Range Dependencies
  4. Causal Discovery and Representation Learning
  5. Causality-Informed Rumor Detection
  6. Graph Neural Networks for Propagation Structure Modeling
  7. Research Gaps
  8. Dataset
  9. Method
  10. Problem Formulation
  11. Sequence Encoding via Mamba
  12. Structural Encoding via GCN
  13. Feature Fusion and Pooling
  14. Rumor Classification
  15. Causal Graph Discovery
  16. ...and 27 more sections

Figures (2)

  • Figure 1: Overview of CausalMamba. Given an input propagation chain, node features are encoded via Mamba and GCN encoders. The fused representation is used for rumor classification and causal graph discovery, optimized jointly via a multi-task loss.
  • Figure 2: Causal graph visualization before and after node intervention. Top-3 nodes (in red) are identified via PageRank. Their removal significantly disrupts information flow, illustrating the model’s interpretability and intervention capability.