Research map / Image generation / Fast diffusion sampling
Decoupling User Relationships Guides Information Diffusion Prediction (Student Abstract)
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 MHGFormer, a multi-channel hypergraph transformer that decouples user relations via triangular motifs into three hypergraphs for information diffusion prediction.
Problem
Information diffusion prediction is limited because user relationships behind resharing are complex, entangled, and high-order, while prior GNN methods only exploit pairwise relations.
Approach
Design triangular motifs to split user relations into society-aware, interest-aware, and joint-aware hypergraphs, then apply a position-aware hypergraph transformer with relation-aware self-gating to learn disentangled user representations for diffusion prediction.
Key results
- On Twitter, MHGFormer reaches H@100 of 58.99 and M@100 of 22.98, beating the best baseline (DyHGCN) by 0.78% and 2.36% respectively.
- On Weibo, MHGFormer reaches H@100 of 55.50 and M@100 of 8.93, the best among compared methods (DyHGCN 51.65 / 8.27 on Weibo).
- Performance decreases more slowly than FOREST, DyHGCN, and MS-HGAT as cascade length grows from 50 to 500.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use MHGFomer instead of DyHGCN | Diffusion prediction model · Twitter | H@100: 58.53 -> 58.99; 0.78% | Helped |
| Use MHGFomer instead of DyHGCN | Diffusion prediction model · Twitter | M@100: 22.45 -> 22.98; 2.36% | Helped |
| Use MHGFomer instead of FOREST | Diffusion prediction model · Weibo | H@100: 52.55 -> 55.50 | Helped |
| Use MHGFomer instead of FOREST | Diffusion prediction model · Weibo | M@100: 8.63 -> 8.93 | Helped |
| Increase cascade length from 50 to 500 | MHGFomer · Twitter and Weibo | MAP@100 | Hurt |
Limitations
- Only two datasets evaluated
- Short student abstract with limited experimental detail
- Results on Twitter gain over best baseline are modest for Hits@100
Experimental setup
Datasets: Twitter Weibo
Benchmarks: Twitter Weibo
Metrics: MAP@100 Hits@100
Baselines: SNIDSA FOREST Inf-VAE DyHGCN MS-HGAT
Builds on: Motif DyHGCN MS-HGAT
Closest papers by meaning
- THGNets: Constrained Temporal Hypergraphs and Graph Neural Networks in Hyperbolic Space for Information Diffusion Prediction (AAAI 2025 · similarity 0.88)
- Enhancing Multi-Scale Diffusion Prediction via Sequential Hypergraphs and Adversarial Learning (AAAI 2024 · similarity 0.88)
- HYGENE: A Diffusion-Based Hypergraph Generation Method (AAAI 2025 · similarity 0.86)
- MSR: A Multifaceted Self-Retrieval Framework for Microscopic Cascade Prediction (AAAI 2025 · similarity 0.84)
- THGFormer: Time-Aware Hypergraph Learning for Multimodal Social Media Popularity Prediction (Student Abstract) (AAAI 2024 · similarity 0.84)
- Learning Multiple User Distributions for Recommendation via Guided Conditional Diffusion (AAAI 2025 · similarity 0.84)
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.