Research map / Graph learning / GNN architectures
Message Passing Attention Networks for Document Understanding
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use 4 rather than 2 message-passing iterations (ablation) | MPAD · Reuters | accuracy: 97.07 -> 97.48 | Helped |
| Use 4 rather than 2 message-passing iterations (ablation) | MPAD · Polarity | accuracy: 80.24 -> 80.52 | Helped |
| Use 4 rather than 2 message-passing iterations (ablation) | MPAD · IMDB | accuracy: 91.30 -> 91.30 | No clear effect |
| Make word co-occurrence edges undirected (ablation) | MPAD · Reuters | accuracy: 97.07 -> 97.35 | Helped |
| Make word co-occurrence edges undirected (ablation) | MPAD · Polarity and IMDB | accuracy: 80.24 -> 80.05; 91.30 -> 90.97 | Hurt |
| Add a master node (ablation) | MPAD · Reuters, Polarity, IMDB | accuracy: 96.66 -> 97.07; 79.15 -> 80.24; 91.09 -> 91.30 | Helped |
| Use weighted averaging rather than weighted summation of neighbor features (ablation) | MPAD · Reuters, Polarity, IMDB | accuracy: 96.02 -> 97.07; 79.84 -> 80.24; 91.16 -> 91.30 | Helped |
| Retain the GRU update of the node's own representation (ablation) | MPAD · Reuters, Polarity, IMDB | accuracy: 97.12 -> 97.07; 79.22 -> 80.24; 89.50 -> 91.30 | Mixed |
| Add the master-node skip connection (ablation) | MPAD · Reuters and IMDB | accuracy: 96.93 -> 97.07; 91.12 -> 91.30 | Helped |
| Add the master-node skip connection (ablation) | MPAD · Polarity | accuracy: 80.62 -> 80.24 | Hurt |
| Use hierarchical sentence-level encoding | MPAD · 9 of 10 classification datasets | accuracy | Helped |
| Use hierarchical sentence-level encoding | MPAD · Subjectivity | accuracy: 93.46* | Hurt |
Closest papers by meaning
- Hierarchical Heterogeneous Graph Attention Network for Syntax-Aware Summarization (AAAI 2022 · similarity 0.86)
- Attention Mechanisms Perspective: Exploring LLM Processing of Graph-Structured Data (ICML 2025 · similarity 0.84)
- Dependency Structure-Enhanced Graph Attention Networks for Event Detection (AAAI 2024 · similarity 0.83)
- Towards Dynamic Message Passing on Graphs (NeurIPS 2024 · similarity 0.83)
- Unifying Text Semantics and Graph Structures for Temporal Text-attributed Graphs with Large Language Models (NeurIPS 2025 · similarity 0.83)
- Training-Free Message Passing for Learning on Hypergraphs (ICLR 2025 · similarity 0.83)
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.