Research map / Graph learning / Graph self-supervised learning
An Empirical Study Towards Prompt-Tuning for Graph Contrastive Pre-Training in Recommendations
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 |
|---|---|---|---|
| add prompt-tuning on GCL pre-trained embeddings (CPTPP-M vs best baseline) | CPTPP · Douban | Precision@5: 0.2187 -> 0.2235 | Helped |
| add prompt-tuning (CPTPP-H vs NCL) | CPTPP · ML-1M | NDCG@5: 0.2818 -> 0.2919 | Helped |
| add prompt-tuning (CPTPP-R vs SimGCL) | CPTPP · Gowalla | Hit Ratio@20: 0.1065 -> 0.1120 | Helped |
| add prompt-tuning (CPTPP-R vs SimGCL) | CPTPP · Gowalla | NDCG@20: 0.1126 -> 0.1092 | No clear effect |
| replace BPR-MF with contrastive-learning pretraining baselines | BPR-MF · Douban | NDCG@5: 0.1904 -> 0.2313 | Helped |
| use GNN-based GCL (NCL/SimGCL) instead of non-graph SSL (BUIR/SelfCF) | recommender · ML-1M | NDCG@5: 0.2689 -> 0.2818 | Helped |
| generate prompts via matrix factorization (CPTPP-M) instead of historical interactions on large-user datasets (ablation) | CPTPP · Gowalla | Precision@5: 0.0755 -> 0.0745 | Hurt |
| generate prompts via matrix factorization (CPTPP-M) on small-user dataset Douban vs CPTPP-H (ablation) | CPTPP · Douban | Precision@5: 0.2221 -> 0.2235 | Helped |
| use high-order user relations for prompts (CPTPP-R) vs historical interactions (CPTPP-H) (ablation) | CPTPP · Gowalla | Precision@5: 0.0754 -> 0.0755 | Helped |
| increase prompt size above the embedding dimension 64 (ablation) | CPTPP · Douban, ML-1M, Gowalla | Precision@5 | Hurt |
| use prompt size 256 (outlier) (ablation) | CPTPP · ML-1M, Gowalla | Precision@5 | No clear effect |
Closest papers by meaning
- DivGCL: A Graph Contrastive Learning Model for Diverse Recommendation (AAAI 2025 · similarity 0.87)
- DisCo: Graph-Based Disentangled Contrastive Learning for Cold-Start Cross-Domain Recommendation (AAAI 2025 · similarity 0.86)
- One Prompt Fits All: Universal Graph Adaptation for Pretrained Models (NeurIPS 2025 · similarity 0.86)
- Edge Prompt Tuning for Graph Neural Networks (ICLR 2025 · similarity 0.86)
- HeterGP: Bridging Heterogeneity in Graph Neural Networks with Multi-View Prompting (AAAI 2025 · similarity 0.86)
- Does Graph Prompt Work? A Data Operation Perspective with Theoretical Analysis (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.