Research map / Graph learning / Graph algorithms and clustering
Graph-Preference Learning: Debiasing Network-Sampled Human Feedback for Target Welfare Estimation
Guangrui Fan, DanDan Liu, AZNUL QALID MD SABRI, Pan Lihu
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 |
|---|---|---|---|
| add graph-balanced aggregation (GBA) to GPRM (ablation) | GPRM · LMArena | pairwise accuracy: 54.3 -> 63.6 | Improved |
| add GBA to GPRM (ablation) | GPRM · LMArena | cross-language accuracy gap: 45.3% -> 38.4% | Improved |
| add GBA to GPRM (ablation) | GPRM · SBM community-biased | welfare MSE: 0.516 -> 0.220 | Improved |
| add GBA to GPRM (ablation) | GPRM · SBM degree-biased | welfare MSE vs degree bias: 0.195 -> 0.207; degree bias 0.900 -> 0.354 | Mixed |
| add graph-personalized reward model (GPRM) alone to standard reward model (ablation) | Standard RM · LMArena | accuracy and log loss: accuracy 55.8 -> 54.3; log loss 1.110 -> 0.692 | Mixed |
| add GPRM+GBA to standard reward model | Standard RM · LMArena | pairwise accuracy: 55.8 -> 63.6 | Improved |
| add GPRM+GBA to standard reward model | Standard RM · LMArena | ECE: 0.323 -> 0.119 | Improved |
| add GPRM+GBA on scale-free BA graph under degree bias | Standard RM · BA degree-biased | welfare MSE: 0.123 -> 0.079 | Improved |
| add GPRM alone on WS graph (ablation) | Standard RM · WS degree-biased | welfare MSE: 0.083 -> 0.090 | Hurt |
| add group-based reweighting | Standard RM · LMArena | pairwise accuracy: 55.8 -> 52.9 | Hurt |
| add DRO-CVaR | Standard RM · SBM uniform | welfare MSE: 0.123 -> 0.771 | Hurt |
| add GPRM+GBA under uniform sampling (q = pi) | Standard RM · SBM uniform | welfare MSE: 0.123 -> 0.080 | Improved |
Closest papers by meaning
- Distributional Preference Learning: Understanding and Accounting for Hidden Context in RLHF (ICLR 2024 · similarity 0.89)
- Capturing Individual Human Preferences with Reward Features (NeurIPS 2025 · similarity 0.89)
- ResponseRank: Data-Efficient Reward Modeling through Preference Strength Learning (NeurIPS 2025 · similarity 0.89)
- PILAF: Optimal Human Preference Sampling for Reward Modeling (ICML 2025 · similarity 0.88)
- Preference Learning with Response Time: Robust Losses and Guarantees (NeurIPS 2025 · similarity 0.88)
- Data with High and Consistent Preference Difference Are Better for Reward Model (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.
