Evidence Retrieval on 2Wiki
40.7Recall@5INTRA
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| INTRA2026.05 | 40.7 | 50.3 | 55.2 | |
| Qwen3-Emb-4B + Jina rerankerparameters=4B, reranker=Jina2026.05 | 35.4 | 40.3 | 43.5 | |
| Qwen3-Emb-4Bparameters=4B2026.05 | 32.8 | 37.2 | 40.8 | |
| BGEvariant=BGE-large2026.05 | 30.9 | 35.9 | 40.1 | |
| Hybrid RAGfusion=reciprocal rank fusion (RRF)2026.05 | 29.1 | 36 | 40.9 | |
| Qwen3-Emb-0.6Bparameters=0.6B2026.05 | 28 | 33.3 | 36.8 | |
| BM25type=sparse lexical2026.05 | 17.4 | 23.2 | 28.4 | |
| MaxSimtype=late-interaction, style=ColBERT2026.05 | 16.7 | 22.6 | 27.5 | |
| TF-IDFtype=sparse lexical2026.05 | 14 | 19.3 | 24.7 |