Research map / Multi-agent systems and games / Human-AI and goal-conditioned
HHAN: Comprehensive Infectious Disease Source Tracing via Heterogeneous Hypergraph Neural Network
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes HHAN, a heterogeneous hypergraph attention network combined with agent-based SIR modeling to trace infection sources in heterogeneous populations, evaluated on three contact datasets against SLVAE, DDMIX, LPSI and CULT.
Problem
Tracing infection sources in heterogeneous populations where transmission differs across media and groups and community structures (hyperedges) drive nonlinear spread, which homogeneous-network source detection methods overlook.
Approach
Models hyperedges as a separate infectious node type C, builds a heterogeneous graph of A (contaminant), B (human) and C (community) nodes, and combines GAT-based heterogeneous GNN layers with an agent-based SIR simulation module and engineered lagged/graph features to classify early-infected nodes.
Key results
- On the Conference dataset HHAN reaches accuracy 0.981 and F1 0.883 (Table 2), versus CULT at 0.887 accuracy and F1 0.063.
- On the Hospital dataset HHAN reports accuracy 0.955 and AUC 0.809, compared with DDMIX/SLVAE AUC around 0.50.
- Ablation on Conference shows removing ABM (HHAN2) drops F1 from 0.883 to 0.809 and replacing GAT with GCN (HHCN) drops F1 to 0.776.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use HHAN instead of CULT | Infectious disease source tracing · Hospital | Accuracy: 0.9087 -> 0.955 | Helped |
| Use HHAN instead of CULT | Infectious disease source tracing · Conference | Accuracy: 0.887 -> 0.981 | Helped |
| Use HHAN instead of CULT | Infectious disease source tracing · School | Accuracy: 0.940 -> 0.765 | Hurt |
| Add feature engineering instead of manual feature selection (ablation) | HHAN on Conference · Conference | Accuracy: 0.871 -> 0.981 | Helped |
| Add Graph Attention Network instead of Graph Convolutional Network (ablation) | HHAN on Conference · Conference | Accuracy: 0.921 -> 0.981 | Helped |
| Add the ABM component (ablation) | HHAN on Conference · Conference | Accuracy: 0.935 -> 0.981 | Helped |
Surprising findings
- The LPSI baseline has precision 1.000 but very low recall, illustrating the imbalance of baselines.
- Table 2 HHAN Conference precision (0.977) and Table 3 values are inconsistent in places, and the School result is reported with lower accuracy (0.765) than the baselines' accuracy (0.940 for CULT).
Limitations
- Evaluated only on SIR-simulated diffusion over three small contact networks (about 75 to 110 nodes).
- Several reported metrics are inconsistent across tables and the School accuracy is lower than baselines.
- The ABM component and hyperedge community detection are described only at a high level, limiting reproducibility.
Experimental setup
Datasets: ACM Hypertext Conference contact dataset School contact dataset (Marseille) Hospital contact dataset (Lyon)
Benchmarks: ACM Hypertext Conference contact dataset (SIR simulated) School contact dataset (SIR simulated) Hospital contact dataset (SIR simulated)
Metrics: accuracy precision recall F1 score ROC-AUC
Baselines: SLVAE DDMIX LPSI CULT HHCN (GCN variant)
Builds on: Graph Attention Network SIR model agent-based modeling DDMIX SLVAE
Closest papers by meaning
- A Unifying Variational Inference Framework for Hierarchical Graph-Coupled HMM with an Application to Influenza Infection (AAAI 2016 · similarity 0.83)
- EARTH: Epidemiology-Aware Neural ODE with Continuous Disease Transmission Graph (ICML 2025 · similarity 0.82)
- Learning Diffusions under Uncertainty (AAAI 2024 · similarity 0.82)
- Modelling Microbial Communities with Graph Neural Networks (ICML 2024 · similarity 0.81)
- When Hypergraph Meets Heterophily: New Benchmark Datasets and Baseline (AAAI 2025 · similarity 0.81)
- A Thorough Comparison Between Independent Cascade and Susceptible-Infected-Recovered Models (AAAI 2025 · similarity 0.80)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.