Document Retrieval on Russian-Wikipedia corpus 10^6 documents
0.834Acc@1Baseline_proj (SVD k)
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Baseline_proj (SVD k)Encoder=multilingual-e5-base, Original Dimension (d)=768, Projected Dimension (k)=384, Subspaces (M)=96, sigma_rec=0.1292026.06 | 0.834 | 0.924 | 0.865 | 0.879 | — | — | |
| ProposedEncoder=multilingual-e5-base, Original Dimension (d)=768, Projected Dimension (k)=384, Subspaces (M)=96, sigma_rec=0.1292026.06 | 0.833 | 0.924 | 0.864 | 0.879 | 0.003 | 356 | |
| Baseline_proj (SVD k)Encoder=BAAI/bge-m3, Original Dimension (d)=1024, Projected Dimension (k)=512, Subspaces (M)=128, sigma_rec=0.1282026.06 | 0.832 | 0.926 | 0.863 | 0.878 | — | — | |
| ProposedEncoder=BAAI/bge-m3, Original Dimension (d)=1024, Projected Dimension (k)=512, Subspaces (M)=128, sigma_rec=0.1282026.06 | 0.832 | 0.924 | 0.863 | 0.878 | 0.014 | 221 | |
| Baseline (raw d)Encoder=multilingual-e5-base, Original Dimension (d)=768, Projected Dimension (k)=384, Subspaces (M)=96, sigma_rec=0.1292026.06 | 0.83 | 0.934 | 0.862 | 0.88 | — | — | |
| Baseline (raw d)Encoder=BAAI/bge-m3, Original Dimension (d)=1024, Projected Dimension (k)=512, Subspaces (M)=128, sigma_rec=0.1282026.06 | 0.818 | 0.908 | 0.848 | 0.862 | — | — | |
| Baseline_proj (SVD k)Encoder=multilingual-e5-large, Original Dimension (d)=1024, Projected Dimension (k)=512, Subspaces (M)=128, sigma_rec=0.1012026.06 | 0.814 | 0.922 | 0.848 | 0.866 | — | — | |
| ProposedEncoder=multilingual-e5-large, Original Dimension (d)=1024, Projected Dimension (k)=512, Subspaces (M)=128, sigma_rec=0.1012026.06 | 0.814 | 0.92 | 0.848 | 0.865 | 0.018 | 219 | |
| Baseline (raw d)Encoder=multilingual-e5-large, Original Dimension (d)=1024, Projected Dimension (k)=512, Subspaces (M)=128, sigma_rec=0.1012026.06 | 0.796 | 0.88 | 0.824 | 0.838 | — | — | |
| Baseline (raw d)Encoder=multilingual-e5-small, Original Dimension (d)=384, Projected Dimension (k)=192, Subspaces (M)=48, sigma_rec=0.2392026.06 | 0.782 | 0.89 | 0.817 | 0.834 | — | — | |
| Baseline (raw d)Encoder=paraphrase-multilingual-mpnet-base-v2, Original Dimension (d)=768, Projected Dimension (k)=384, Subspaces (M)=96, sigma_rec=0.1072026.06 | 0.716 | 0.842 | 0.762 | 0.782 | — | — | |
| Baseline_proj (SVD k)Encoder=multilingual-e5-small, Original Dimension (d)=384, Projected Dimension (k)=192, Subspaces (M)=48, sigma_rec=0.2392026.06 | 0.702 | 0.834 | 0.747 | 0.768 | — | — | |
| ProposedEncoder=multilingual-e5-small, Original Dimension (d)=384, Projected Dimension (k)=192, Subspaces (M)=48, sigma_rec=0.2392026.06 | 0.701 | 0.831 | 0.746 | 0.767 | 0.081 | 314 | |
| Baseline_proj (SVD k)Encoder=paraphrase-multilingual-mpnet-base-v2, Original Dimension (d)=768, Projected Dimension (k)=384, Subspaces (M)=96, sigma_rec=0.1072026.06 | 0.65 | 0.796 | 0.693 | 0.717 | — | — | |
| ProposedEncoder=paraphrase-multilingual-mpnet-base-v2, Original Dimension (d)=768, Projected Dimension (k)=384, Subspaces (M)=96, sigma_rec=0.1072026.06 | 0.649 | 0.795 | 0.692 | 0.717 | 0.067 | 351 |