Research map / Image generation / Fast diffusion sampling
Public Opinion Field Effect and Hawkes Process Join Hands for Information Popularity 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 POFHP, a Hawkes-process-based model that combines an improved neural Hawkes process with a public opinion field paradigm and cascade graph representation learning to predict information popularity.
Problem
Predicting the incremental popularity of trending topics in social networks while accounting for multiple topics coexisting and competing for user attention (the public opinion field effect), which prior single-process diffusion models ignore.
Approach
Extends the neural Hawkes process with intra-topic and inter-topic influence terms in the conditional intensity, defines public opinion fields (topic-centered subgraphs) weighted by field energy from the intensity, and uses these in a heterogeneous cascade graph representation learner that pools field embeddings to predict incremental popularity.
Key results
- POFHP achieves MSLE 1.5197 on Twitter vs 1.7774 for the second-best POFD, a 14.50% improvement.
- On Douban, POFHP reaches MSLE 1.0347 (16.47% improvement over CTCP); on Christianity, MSLE 0.3906 (35.38% improvement over CTCP).
- Ablation shows removing the Hawkes process (w/o HP) raises Twitter MSLE from 1.5197 to 1.7012.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use POFHP instead of POFD | Information popularity prediction · Twitter | MSLE: 1.7774 -> 1.5197 | Helped |
| use POFHP instead of POFD | Information popularity prediction · Douban | MSLE: 1.2387 -> 1.0347 | Helped |
| use POFHP instead of POFD | Information popularity prediction · Android | MSLE: 1.2834 -> 0.6986 | Helped |
| use POFHP instead of POFD | Information popularity prediction · Christianity | MSLE: 0.6045 -> 0.3906 | Helped |
| add the Hawkes Process and its Lm loss (ablation) | POFHP · Twitter | MSLE: 1.7012 -> 1.5197 | Helped |
| add short-term dependencies (ablation) | POFHP · Twitter | MSLE: 1.6113 -> 1.5197 | Helped |
| add long-term dependencies (ablation) | POFHP · Twitter | MSLE: 1.6514 -> 1.5197 | Helped |
| add the readout function (ablation) | POFHP · Twitter | MSLE: 1.5438 -> 1.5197 | Helped |
| add the Hawkes Process and its Lm loss (ablation) | POFHP · Douban | MSLE: 1.3258 -> 1.0347 | Helped |
| add short-term dependencies (ablation) | POFHP · Douban | MSLE: 1.2478 -> 1.0347 | Helped |
| add long-term dependencies (ablation) | POFHP · Douban | MSLE: 1.2784 -> 1.0347 | Helped |
| use the readout function instead of a two-layer MLP (ablation) | POFHP · Douban | MSLE: 1.1517 -> 1.0347 | Helped |
Limitations
- Evaluated only on four datasets, two of which are small (Android, Christianity).
- Model relies on heterogeneous cascade graph construction with friendship networks, which may not be available in all settings.
- Compared against baselines with fixed hyperparameters; sensitivity analysis limited to Twitter.
Experimental setup
Datasets: Twitter Douban Android Christianity
Benchmarks: Twitter Douban Android Christianity
Metrics: MSLE MALE SMAPE
Baselines: XGBoost MLP SEISMIC DeepHawkes CasCN MS-HGAT CTCP POFD
Builds on: neural Hawkes process POFD graph attention network Transformer CasCN
Closest papers by meaning
- CasFT: Future Trend Modeling for Information Popularity Prediction with Dynamic Cues-Driven Diffusion Models (AAAI 2025 · similarity 0.88)
- Enhancing Multi-Scale Diffusion Prediction via Sequential Hypergraphs and Adversarial Learning (AAAI 2024 · similarity 0.84)
- SMTPD: A New Benchmark for Temporal Prediction of Social Media Popularity (CVPR 2025 · similarity 0.84)
- Improving Multimodal Social Media Popularity Prediction via Selective Retrieval Knowledge Augmentation (AAAI 2025 · similarity 0.83)
- Opinion Maximization in Social Networks by Modifying Internal Opinions (NeurIPS 2025 · similarity 0.82)
- Ghidorah: Towards Robust Multi-Scale Information Diffusion Prediction via Test-Time Training (AAAI 2025 · 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.