Research map / Graph learning / Graph algorithms and clustering
BlitzRank: Principled Zero-shot Ranking Agents with Tournament Graphs
Sheshansh Agrawal, Thien Hang Nguyen, Douwe Kiela
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 |
|---|---|---|---|
| use greedy SCC scheduling instead of random candidate subsets (ablation) | BlitzRank · Synthetic transitive n=100,k=10 | oracle calls: 132.1 -> 37.7 | Improved |
| use greedy SCC scheduling instead of random subsets from all nodes (ablation) | BlitzRank · Synthetic transitive n=200,k=10 | oracle calls: 2554.6 -> 89.2 | Improved |
| use greedy SCC scheduling instead of random candidates (ablation) | BlitzRank · Synthetic transitive n=200,k=20 | oracle calls: 129.4 -> 33.8 | Improved |
| reduce window size from 20 to 10 | Sliding Window · TREC DL19 (GPT-4.1) | nDCG@10: 74.0 -> 56.4 | Hurt |
| reduce window size from 20 to 10 | BlitzRank · TREC DL19 (GPT-4.1) | nDCG@10: 74.6 -> 73.6 | No clear effect |
| reduce window size from 20 to 10 | BlitzRank · TREC DL20 (GPT-4.1) | nDCG@10: 70.7 -> 72.4 | No clear effect |
| use BlitzRank k10 instead of k20 | BlitzRank · 14 datasets x 5 LLMs avg | nDCG@10: 56.0 -> 56.5 | Improved |
| use BlitzRank k10 instead of k20 | BlitzRank · Qwen3-235B | nDCG@10: 54.7 -> 55.6 | Improved |
| use BlitzRank k10 instead of pairwise reranking | LLM reranking · 14 datasets (GPT-4.1 & Gemini-3-Flash) | tokens per query: 324k -> 42k | Improved |
| use BlitzRank k10 instead of Sliding Window | LLM reranking · 14 datasets (GPT-4.1 & Gemini-3-Flash) | tokens per query: 54k -> 42k | Improved |
| use BlitzRank k10 instead of AcuRank | LLM reranking · 14 datasets (GPT-4.1 & Gemini-3-Flash) | tokens per query: 69k -> 42k | Improved |
| double compute with repeated runs (R2 variant) | SW · 14 datasets (GPT-4.1 & Gemini-3-Flash) | nDCG@10: 56.7 -> 56.5 | No clear effect |
Closest papers by meaning
- TopRank: A practical algorithm for online stochastic ranking (NeurIPS 2018 · similarity 0.87)
- Active Ranking and Matchmaking, with Perfect Matchings (ICML 2024 · similarity 0.87)
- Dynamic Ranking with the BTL Model: A Nearest Neighbor based Rank Centrality Method (JMLR 2023 · similarity 0.87)
- Maximum Selection and Ranking under Noisy Comparisons (ICML 2017 · similarity 0.87)
- Eliciting Kemeny Rankings (AAAI 2024 · similarity 0.85)
- Ranking with Multiple Oracles: From Weak to Strong Stochastic Transitivity (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.
