Query–Fragment Retrieval on Multi-modal dialogue corpus
62.8Recall@1FEM
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| FEMZero-shot inference=false2026.06 | 62.8 | 89.6 | 94.2 | 97.4 | 98.6 | |
| GME-Qwen2-VL w/ Inter-Fragment ContrastiveInter-Fragment Contrastive=true, Zero-shot inference=false2026.06 | 58 | 87.8 | 92.4 | 96 | 97.2 | |
| GME-Qwen2-VL w/ Intra-Fragment ContrastiveIntra-Fragment Contrastive=true, Zero-shot inference=false2026.06 | 52.6 | 83.4 | 91 | 93.8 | 95.8 | |
| GME-Qwen2-VL(2B)Zero-shot inference=true, Query–Fragment data fine-tuning=false2026.06 | 38.8 | 65 | 74.4 | 79.4 | 83.4 | |
| VLM2Vec(7B)Zero-shot inference=true, Query–Fragment data fine-tuning=false2026.06 | 29.8 | 47.8 | 54.5 | 62.1 | 67.5 | |
| LamRA(7B)Zero-shot inference=true, Query–Fragment data fine-tuning=false2026.06 | 28.2 | 47.6 | 53.2 | 61.3 | 66.8 | |
| MonoQwen-ReRanker(2B)Zero-shot inference=true, Query–Fragment data fine-tuning=false2026.06 | 27.6 | 56 | 70.6 | 76.8 | 80.4 | |
| Jina-embeddings-v4(3B)Zero-shot inference=true, Query–Fragment data fine-tuning=false2026.06 | 21.2 | 48.8 | 57.6 | 64.3 | 69 |