Research map / Graph learning / Graph self-supervised learning
LightGCL: Simple Yet Effective Graph Contrastive Learning for Recommendation
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 SVD-based global graph augmentation and local-global contrastive learning instead of SimGCL's embedding-noise augmentation | LightGCL · Yelp | Recall@20: 0.0718 -> 0.0793 | Helped |
| Use SVD-based global graph augmentation and local-global contrastive learning instead of SimGCL's embedding-noise augmentation | LightGCL · Gowalla | Recall@20: 0.1357 -> 0.1578 | Helped |
| Use SVD-based global graph augmentation and local-global contrastive learning instead of SimGCL's embedding-noise augmentation | LightGCL · ML-10M | Recall@20: 0.2265 -> 0.2613 | Helped |
| Use SVD-based global graph augmentation and local-global contrastive learning instead of SimGCL's embedding-noise augmentation | LightGCL · Amazon-book | Recall@20: 0.0474 -> 0.0585 | Helped |
| Use SVD-based global graph augmentation and local-global contrastive learning instead of SimGCL's embedding-noise augmentation | LightGCL · Tmall | Recall@20: 0.0473 -> 0.0528 | Helped |
| Replace approximate SVD view with a view from pre-trained matrix factorization (ablation) | LightGCL · Datasets in ablation study | Recommendation performance | Hurt |
| Replace approximate SVD view with an SVD++ view (ablation) | LightGCL · Datasets in ablation study | Recommendation performance | Hurt |
| Use the new positive/negative item-pair sampling setting rather than the original user-first sampling setting (ablation) | LightGCL · Yelp | Recall@20: 0.1050 vs 0.0985 | Mixed |
| Use the new positive/negative item-pair sampling setting rather than the original user-first sampling setting (ablation) | LightGCL · Amazon-book | Recall@20: 0.1116 vs 0.1161 | Helped |
| Set the SVD rank to q = 5 (ablation) | LightGCL · User-item interaction graph | Recommendation performance | No clear effect |
| Vary temperature across the reported range (ablation) | LightGCL · Yelp, Gowalla, and ML-10M | Recommendation performance | No clear effect |
Closest papers by meaning
- DivGCL: A Graph Contrastive Learning Model for Diverse Recommendation (AAAI 2025 · similarity 0.91)
- Robust Graph Based Social Recommendation Through Contrastive Multi-View Learning (AAAI 2025 · similarity 0.89)
- Review-Enhanced Hierarchical Contrastive Learning for Recommendation (AAAI 2024 · similarity 0.88)
- SHARP-Distill: A 68× Faster Recommender System with Hypergraph Neural Networks and Language Models (ICML 2025 · similarity 0.87)
- Temporal Graph Contrastive Learning for Sequential Recommendation (AAAI 2024 · similarity 0.87)
- Graph Disentangled Contrastive Learning with Personalized Transfer for Cross-Domain Recommendation (AAAI 2024 · similarity 0.86)
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.