Research map / Large language models / Language, safety and interpretability
Whom to Query for What: Adaptive Group Elicitation via Multi-Turn LLM Interactions
Ruomeng Ding, Tianwei Gao, Thomas P Zollo, Eitan Bachmat, Richard Zemel, Zhun Deng
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 |
|---|---|---|---|
| jointly select questions and respondents with LLM EIG plus GNN imputation versus Meta baselines | Ours vs Meta-Random / Meta-Greedy / Meta-Greedy-Imp · CES, OpinionQA, Twin-2K at 10/30/50% budgets | accuracy on held-out target questions: 17.1% relative at round 1 to 12.6% by round 4 (CES, 10% budget) | Improved |
| add LLM-based imputation to Meta-Greedy | Meta-Greedy · CES, OpinionQA, Twin-2K | accuracy | No clear effect |
| greedy EIG query selection instead of random query selection (both with imputation) (ablation) | GreedyQ-Imp vs RandomQ-Imp · CES and OpinionQA | accuracy | Improved |
| add GNN-based imputation (ablation) | GreedyQ · CES and OpinionQA | accuracy | Improved |
| actively observe selected respondents on top of imputation-only (ablation) | Ours-Imp · CES and OpinionQA | accuracy | Improved |
| group-relational respondent selection instead of random respondent selection (ablation) | Ours · CES and OpinionQA, hard respondents | relative recovery: up to 20% relative recovery under a 50% budget | Improved |
| multi-step planning instead of greedy query selection (ablation) | Greedy selection · Round 4, 10% budget, hard respondents | accuracy: 0.356 -> 0.372 | Improved |
| multi-step planning instead of greedy query selection (ablation) | Greedy selection · Round 4, 50% budget, extreme respondents (top 5%) | accuracy: 0.561 -> 0.542 | Hurt |
| multi-step planning instead of greedy query selection (ablation) | Greedy selection · Round 4, global population | accuracy: 0.488 -> 0.485 (10% budget) | No clear effect |
Closest papers by meaning
- More Is Not More: What Matters for Diversity in LLM Opinions? (NeurIPS 2026 · similarity 0.88)
- Adaptive Elicitation of Latent Information Using Natural Language (ICML 2025 · similarity 0.88)
- Prompts to Proxies: Emulating Human Preferences via a Compact LLM Ensemble (NeurIPS 2026 · similarity 0.88)
- Questioning the Survey Responses of Large Language Models (NeurIPS 2024 · similarity 0.87)
- DEBATE: A Large-Scale Benchmark for Evaluating Opinion Dynamics in Role-Playing LLM Agents (NeurIPS 2026 · similarity 0.87)
- Examining Alignment of Large Language Models through Representative Heuristics: the case of political stereotypes (ICLR 2025 · 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.
