Research map / Large language models / Retrieval and knowledge
GNN is a Counter? Revisiting GNN for Question Answering
Kuan Wang, Yuyu Zhang, Diyi Yang, Le Song, Tao Qin
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 |
|---|---|---|---|
| replace complex graph reasoning with Graph Soft Counter | RoBERTa-large QA with ConceptNet · CommonsenseQA IHdev | accuracy: 76.54~($\pm$0.21) -> 79.11~($\pm$0.22) | Improved |
| replace complex graph reasoning with Graph Soft Counter | RoBERTa-large QA with ConceptNet · CommonsenseQA IHtest | accuracy: 73.41~($\pm$0.92) -> 74.48~($\pm$0.41) | Improved |
| replace complex graph reasoning with Graph Soft Counter | RoBERTa-large QA with ConceptNet · OpenBookQA test | accuracy: 67.80\footnotemark~($\pm$2.75) -> 70.33~($\pm$0.81) | Improved |
| replace complex graph reasoning with Graph Soft Counter | AristoRoBERTa QA with ConceptNet · OpenBookQA test | accuracy: 82.77~($\pm$1.56) -> 86.67~($\pm$0.46) | Improved |
| use hard 2-hop edge counts with a two-layer MLP instead of soft graph counting (ablation) | GSC on CommonsenseQA · CommonsenseQA IHtest | accuracy: 74.48~($\pm$0.41) -> 74.13~($\pm$0.08) | No clear effect |
| use hard 1-hop edge counts with a two-layer MLP instead of soft graph counting (ablation) | GSC on CommonsenseQA · CommonsenseQA IHtest | accuracy: 74.48~($\pm$0.41) -> 73.62~($\pm$0.12) | Hurt |
| increase retrieved-node cap from QA-context entities to 32 (ablation) | GSC on CommonsenseQA · CommonsenseQA IHtest | accuracy: 74.48~($\pm$0.41) -> 74.40~($\pm$0.25) | No clear effect |
| increase retrieved-node cap from QA-context entities to 64 (ablation) | GSC on CommonsenseQA · CommonsenseQA IHtest | accuracy: 74.48~($\pm$0.41) -> 73.93~($\pm$1.09) | Hurt |
| increase retrieved-node cap from QA-context entities to 128 (ablation) | GSC on CommonsenseQA · CommonsenseQA IHtest | accuracy: 74.48~($\pm$0.41) -> 72.78~($\pm$1.15) | Hurt |
| increase retrieved-node cap from QA-context entities to 256 (ablation) | GSC on CommonsenseQA · CommonsenseQA IHtest | accuracy: 74.48~($\pm$0.41) -> 73.89~($\pm$0.63) | Hurt |
| apply SparseVD pruning during training (ablation) | KagNet · CommonsenseQA IHtest | accuracy: 69.01~($\pm$0.76) -> 70.48~($\pm$0.77) | Improved |
| apply SparseVD pruning during training (ablation) | MHGRN · CommonsenseQA IHtest | accuracy: 71.11~($\pm$0.81) -> 72.66~($\pm$0.61) | Improved |
Closest papers by meaning
- Relation-Aware Language-Graph Transformer for Question Answering (AAAI 2023 · similarity 0.88)
- Debate on Graph: A Flexible and Reliable Reasoning Framework for Large Language Models (AAAI 2025 · similarity 0.87)
- Graph Reasoning Transformers for Knowledge-Aware Question Answering (AAAI 2024 · similarity 0.86)
- KG-Guard: Graph-Based Hallucination Detection for Knowledge Base Question Answering (NeurIPS 2026 · similarity 0.85)
- Question Calibration and Multi-Hop Modeling for Temporal Question Answering (AAAI 2024 · similarity 0.85)
- Graph-constrained Reasoning: Faithful Reasoning on Knowledge Graphs with Large Language Models (ICML 2025 · similarity 0.85)
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.
