Research map / Graph learning / Knowledge and temporal graphs
GKD-Recruiter: Jointly Modeling Social and Task Heterogeneity for Spatial Crowdsourcing via Graph Knowledge Distillation
Yucen Gao, Zhemeng Yu, Zhuoran Li, Jianxiong Guo, Xiaofeng Gao
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 the RGCN worker-task module (ablation) | GKD-Recruiter · Gowalla (|V|=3000, K=100) | ETS: 0.2730 -> 0.2888 | Improved |
| use Influential GAT instead of standard GAT (ablation) | GKD-Recruiter · Gowalla (|V|=3000, K=100) | ETS: 0.2760 -> 0.2888 | Improved |
| add the correlation layer (ablation) | GKD-Recruiter · Gowalla (|V|=3000, K=100) | ETS: 0.2820 -> 0.2888 | Improved |
| add the graph knowledge distillation loss (ablation) | GKD-Recruiter · Gowalla (|V|=5000, K=100) | ETS: 0.2810 -> 0.2883 | Improved |
| use more GNN layers (1 -> 3) (ablation) | GKD-Recruiter · Gowalla (|V|=3000, K=100) | ETS: 0.2888 -> 0.2833 | Hurt |
| vary candidate task size m (3, 7, 9 vs 5) (ablation) | GKD-Recruiter · Gowalla (|V|=3000, K=100) | ETS: 0.2888 at m=5 vs 0.2880/0.2879/0.2884 | No clear effect |
| larger candidate size m increases inference time (ablation) | GKD-Recruiter · Gowalla (|V|=3000) | time cost: Low -> Very High (m=3 to m=9) | Hurt |
| use RL with long-term reward instead of greedy CELF at large budget | GKD-Recruiter vs CELF · Gowalla (|V|=5000, K=100 to 150) | ETS: CELF 0.2683 -> 0.2933 (+9.3%) vs GKD 0.2883 -> 0.4051 (+40.5%) | Improved |
| CELF is on par or slightly ahead at small budgets | CELF vs GKD-Recruiter · Gowalla (K <= 75) | ETS | No clear effect |
| fine-tune only GNN parameters after edge perturbation | GKD-Recruiter · Gowalla (|V|=5000, K=100, 5% edges perturbed) | ETS: 0.2883 -> 0.2840 | No clear effect |
| use RL inference instead of Monte Carlo greedy (CELF) | GKD-Recruiter · Gowalla (|V|=5000, K=150) | inference time (sec.): 38.10 -> 6.196 | Improved |
Closest papers by meaning
- Preference-Aware Task Assignment in Spatial Crowdsourcing (AAAI 2019 · similarity 0.87)
- Gradient-Guided Credit Assignment and Joint Optimization for Dependency-Aware Spatial Crowdsourcing (AAAI 2025 · similarity 0.85)
- MAGNOLIA: Matching Algorithms via GNNs for Online Value-to-go Approximation (ICML 2024 · similarity 0.84)
- Making Better Use of the Crowd: How Crowdsourcing Can Advance Machine Learning Research (JMLR 2018 · similarity 0.82)
- Peer Neighborhood Mechanisms: A Framework for Mechanism Generalization (AAAI 2024 · similarity 0.81)
- DICE: Data Influence Cascade in Decentralized Learning (ICLR 2025 · similarity 0.81)
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.
