Research map / Graph learning / Graph self-supervised learning
GraphFormers: GNN-nested Transformers for Representation Learning on Textual Graph
Junhan Yang, Zheng Liu, Shitao Xiao, Chaozhuo Li, Defu Lian, Sanjay Agrawal, Amit S, Guangzhong Sun, Xing Xie
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| nest GNN layers inside transformer layers instead of cascading GNN after the PLM | UniLM-base text graph encoder · DBLP | P@1: 0.6934 -> 0.7267 | Improved |
| nest GNN layers inside transformers instead of cascading | UniLM-base text graph encoder · Wiki | P@1: 0.3712 -> 0.3952 | Improved |
| nest GNN layers inside transformers instead of cascading | UniLM-base text graph encoder · Product | P@1: 0.7570 -> 0.7786 | Improved |
| add two-stage progressive learning (masked warm-up then original data) (ablation) | GraphFormers · DBLP | P@1: 0.7096 -> 0.7267 | Improved |
| add two-stage progressive learning (ablation) | GraphFormers · Product | P@1: 0.7688 -> 0.7786 | Improved |
| add two-stage progressive learning (ablation) | GraphFormers · Wiki | P@1: 0.3834 -> 0.3952 | Improved |
| use unidirectional graph aggregation (simplified) instead of bidirectional (ablation) | GraphFormers · DBLP | P@1: 0.7225 -> 0.7267 | No clear effect |
| share GNN parameters across layers (ablation) | GraphFormers · DBLP | P@1: 0.7256 -> 0.7267 | No clear effect |
| add learnable position bias in GNN (ablation) | GraphFormers · DBLP | P@1: 0.7276 -> 0.7267 | No clear effect |
| increase number of neighbours from 1 to 5 | GraphFormers · DBLP | P@1: 0.6485 -> 0.7267 | Improved |
| use GAT aggregator instead of Max/Mean/Att pooling in cascaded model | PLM+GNN cascaded · Wiki | P@1: 0.3712 -> 0.3006 | Hurt |
| use GraphFormers instead of previous production system | Bing ads retrieval · Online A/B test | RPM: +1.87% | Improved |
Closest papers by meaning
- A Unified Graph Language Model for Multi-Domain Multi-Task Graph Alignment Instruction Tuning (NeurIPS 2026 · similarity 0.88)
- LLMs as Zero-shot Graph Learners: Alignment of GNN Representations with LLM Token Embeddings (NeurIPS 2024 · similarity 0.88)
- Leveraging Contrastive Learning for Enhanced Node Representations in Tokenized Graph Transformers (NeurIPS 2024 · similarity 0.88)
- Scale-Free Graph-Language Models (ICLR 2025 · similarity 0.88)
- UniGTE: Unified Graph–Text Encoding for Zero-Shot Generalization across Graph Tasks and Domains (NeurIPS 2025 · similarity 0.88)
- Bootstrapping Heterogeneous Graph Representation Learning via Large Language Models: A Generalized Approach (AAAI 2025 · similarity 0.88)
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.
