Information Retrieval on NFCorpus (NDCG@10, NDCG@50, Recall@10, Recall@50)
40.6NDCG@10BAGEL
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| BAGELLLM Backbone=GPT-4o, Ranking Protocol=ER, LLM budget per query=502026.04 | 40.6 | 35.9 | 19 | 27.3 | |
| Pointwise LLMLLM Backbone=GPT-4o, Ranking Protocol=ER, LLM budget per query=502026.04 | 39.1 | 33.5 | 18.2 | 25.3 | |
| BAGELLLM Backbone=GPT-4o, Ranking Protocol=PR, LLM budget per query=502026.04 | 38.9 | 32.8 | 18.2 | 24.5 | |
| BAGELLLM Backbone=Qwen3-14B, Ranking Protocol=ER, LLM budget per query=502026.04 | 38.4 | 33.6 | 18.3 | 26.4 | |
| RankGPTLLM Backbone=GPT-4o, Ranking Method=Listwise, LLM budget per query=502026.04 | 38.4 | 33 | 18.1 | 25.3 | |
| Pointwise LLMLLM Backbone=GPT-4o, Ranking Protocol=PR, LLM budget per query=502026.04 | 37.8 | 32.8 | 17.9 | 25.3 | |
| BAGELLLM Backbone=Qwen3-14B, Ranking Protocol=PR, LLM budget per query=502026.04 | 37.7 | 32.8 | 18 | 25.9 | |
| Pointwise LLMLLM Backbone=Qwen3-14B, Ranking Protocol=PR, LLM budget per query=502026.04 | 37.5 | 32.7 | 17.7 | 25.3 | |
| Pointwise LLMLLM Backbone=Qwen3-14B, Ranking Protocol=ER, LLM budget per query=502026.04 | 37.5 | 32.7 | 17.5 | 25.3 | |
| RankGPTLLM Backbone=Baseline, Ranking Method=Listwise, LLM budget per query=502026.04 | 34.7 | 30.9 | 16.8 | 25.3 | |
| Cross EncoderLLM Backbone=Baseline, LLM budget per query=502026.04 | 33.7 | 30.7 | 15.5 | 25.3 | |
| BM25LLM Backbone=Baseline, LLM budget per query=502026.04 | 31.5 | 27.7 | 15.1 | 21.4 | |
| Dense RetrieverLLM Backbone=Baseline, LLM budget per query=502026.04 | 31.2 | 29 | 15.3 | 25.3 |