Research map / Graph learning / Knowledge and temporal graphs
THGNets: Constrained Temporal Hypergraphs and Graph Neural Networks in Hyperbolic Space for Information Diffusion Prediction
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 THGNets, a hyperbolic (Poincaré ball) framework combining a temporal hypergraph network with time-consistency constraints, a hyperbolic GNN for social networks, and hyperbolic GRUs to predict the next infected user in information cascades.
Problem
Information diffusion prediction (predicting the next infected user) fails to capture hierarchical/power-law structure of cascades and social networks in Euclidean space, causing distorted and unstable user features.
Approach
Embed users in hyperbolic space and combine a hyperbolic temporal hypergraph network (with gated snapshot fusion and a time-consistency distance loss) for dynamic high-order interactions with a hyperbolic GNN for social networks and hyperbolic GRUs for dependencies.
Key results
- THGNets reports Hits@100 of 0.6589 on Christianity versus 0.6250 for MINDS, the strongest baseline.
- Average improvements over MINDS of 9.3% and 7.4% are claimed on the Android dataset (per the text).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use THGNets instead of MINDS | Information diffusion prediction · Android | Hits@10 | Helped |
| Add the time-consistency constraint (HTC) (ablation) | THGNets · Christianity and Memetracker | performance | Helped |
| Replace the temporal preference fusion gate with addition (ablation) | THGNets · Christianity and Memetracker | performance | Hurt |
| Replace the interaction/social relationship fusion gate with addition (ablation) | THGNets · Christianity and Memetracker | performance | Hurt |
| Remove the social network (ablation) | THGNets · Christianity and Memetracker | performance | Hurt |
| Replace the hyperbolic hypergraph neural network with a hyperbolic graph neural network (ablation) | THGNets · Christianity and Memetracker | performance | Hurt |
Limitations
- Only curvature fixed across layers; the authors plan to explore different curvatures per layer in future work.
- Excessive isolated cascade subgraphs reduce performance, per the sensitivity analysis.
Experimental setup
Datasets: Christianity Android Douban Memetracker
Benchmarks: Christianity Android Douban Memetracker
Metrics: MAP@k Hits@k (k=10,50,100)
Baselines: Topo-LSTM NDM SNIDSA Inf-VAE DyHGCN TAN-DRUD FOREST DMT-LIC MINDS
Builds on: MINDS HGCN RotDiff HGRU
Closest papers by meaning
- Enhancing Multi-Scale Diffusion Prediction via Sequential Hypergraphs and Adversarial Learning (AAAI 2024 · similarity 0.90)
- Decoupling User Relationships Guides Information Diffusion Prediction (Student Abstract) (AAAI 2024 · similarity 0.88)
- HYGENE: A Diffusion-Based Hypergraph Generation Method (AAAI 2025 · similarity 0.86)
- THGFormer: Time-Aware Hypergraph Learning for Multimodal Social Media Popularity Prediction (Student Abstract) (AAAI 2024 · similarity 0.86)
- Training-Free Message Passing for Learning on Hypergraphs (ICLR 2025 · similarity 0.86)
- Hypergraph Representation Learning with Hyperlink Random Effects (NeurIPS 2026 · similarity 0.86)
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.