Research map / Label-efficient and robust learning / Recognition and long-tail
THGFormer: Time-Aware Hypergraph Learning for Multimodal Social Media Popularity 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 THGFormer, a time-aware hypergraph Transformer that builds per-UGC hypergraphs from top-K time-relevant instances and uses a time encoder with cross-modal gating to predict social media popularity.
Problem
Multimodal social media popularity prediction struggles because existing methods treat each UGC independently and ignore inter-instance correlations and temporal patterns.
Approach
Retrieve the top-K relevant prior instances before the post time, connect them via attribute-based hyperedges (user, category, topic), and apply a hypergraph Transformer with a functional time encoder and gated cross-modal (visual/textual) attention to produce popularity predictions.
Key results
- On SMPD, THGFormer reaches MSE 3.5673, MAE 1.4138, SRC 0.5741, improving over the best baseline by 7.27% MSE, 4.69% MAE and 3.48% SRC.
- On ICIP, THGFormer reaches MSE 1.5615, MAE 0.8326, SRC 0.4759, improving over the best baseline by 16.66% MSE, 8.83% MAE and 17.77% SRC.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use THGFormer instead of UHAN | MSMPP models · SMPD | MSE: 3.8471 -> 3.5673 | Helped |
| Use THGFormer instead of UHAN | MSMPP models · SMPD | MAE: 1.4833 -> 1.4138 | Helped |
| Use THGFormer instead of UHAN | MSMPP models · SMPD | SRC: 0.5541 -> 0.5741 | Helped |
| Use THGFormer instead of MHF | MSMPP models · ICIP | MSE: 1.8736 -> 1.5615 | Helped |
| Use THGFormer instead of MHF | MSMPP models · ICIP | MAE: 0.9132 -> 0.8326 | Helped |
| Use THGFormer instead of MHF | MSMPP models · ICIP | SRC: 0.4041 -> 0.4759 | Helped |
Limitations
- Evaluated only on two datasets
- Short student abstract with limited ablation and implementation detail
Experimental setup
Datasets: SMPD ICIP
Benchmarks: SMPD ICIP
Metrics: MSE MAE SRC
Baselines: SVR Hyfea DTCN UHAN MHF
Builds on: hypergraph neural networks functional time encoding (Xu et al. 2019) UHAN MHF
Closest papers by meaning
- Improving Multimodal Social Media Popularity Prediction via Selective Retrieval Knowledge Augmentation (AAAI 2025 · similarity 0.90)
- SMTPD: A New Benchmark for Temporal Prediction of Social Media Popularity (CVPR 2025 · similarity 0.87)
- THGNets: Constrained Temporal Hypergraphs and Graph Neural Networks in Hyperbolic Space for Information Diffusion Prediction (AAAI 2025 · similarity 0.86)
- Decoupling User Relationships Guides Information Diffusion Prediction (Student Abstract) (AAAI 2024 · similarity 0.84)
- Hgformer: Hyperbolic Graph Transformer for Collaborative Filtering (ICML 2025 · similarity 0.83)
- Enhancing Multi-Scale Diffusion Prediction via Sequential Hypergraphs and Adversarial Learning (AAAI 2024 · similarity 0.82)
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.