Retrieval on CUAD
91.25RecallM3DOCDEP
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| M3DOCDEPRetrievers=Averaged over BGE, E5, BM25, MM-Embed, Top-k=macro-averaged for {1, 2, 3, 4}, RAG Protocol=Shared corpus-level2026.04 | 91.25 | 87.23 | 89.12 | |
| LumberChunkertop-k=1 ~ 42026.04 | 90.31 | 85.76 | 88 | |
| LumberChunkerRetrievers=Averaged over BGE, E5, BM25, MM-Embed, Top-k=macro-averaged for {1, 2, 3, 4}, RAG Protocol=Shared corpus-level2026.04 | 90.31 | 85.76 | 88 | |
| MultiDocFusiontop-k=1 ~ 42026.04 | 90.21 | 86.51 | 88.19 | |
| MultiDocFusionRetrievers=Averaged over BGE, E5, BM25, MM-Embed, Top-k=macro-averaged for {1, 2, 3, 4}, RAG Protocol=Shared corpus-level2026.04 | 90.21 | 86.51 | 88.19 | |
| Length chunkingtop-k=1 ~ 42026.04 | 90.11 | 85.37 | 87.76 | |
| Length chunkingRetrievers=Averaged over BGE, E5, BM25, MM-Embed, Top-k=macro-averaged for {1, 2, 3, 4}, RAG Protocol=Shared corpus-level2026.04 | 90.11 | 85.37 | 87.76 | |
| Perplexity chunkingtop-k=1 ~ 42026.04 | 88.69 | 83.95 | 86.03 | |
| Perplexity chunkingRetrievers=Averaged over BGE, E5, BM25, MM-Embed, Top-k=macro-averaged for {1, 2, 3, 4}, RAG Protocol=Shared corpus-level2026.04 | 88.69 | 83.95 | 86.03 | |
| Structure-based chunkingtop-k=1 ~ 42026.04 | 88.44 | 83.11 | 85.81 | |
| Structure-based chunkingRetrievers=Averaged over BGE, E5, BM25, MM-Embed, Top-k=macro-averaged for {1, 2, 3, 4}, RAG Protocol=Shared corpus-level2026.04 | 88.44 | 83.11 | 85.81 | |
| Semantic chunkingtop-k=1 ~ 42026.04 | 76.84 | 67.19 | 71.81 | |
| Semantic chunkingRetrievers=Averaged over BGE, E5, BM25, MM-Embed, Top-k=macro-averaged for {1, 2, 3, 4}, RAG Protocol=Shared corpus-level2026.04 | 76.84 | 67.19 | 71.81 |