# Learning to Rank

EViSE splits summary evaluation into faithfulness, coherence and relevance and returns all three scores from a single non-autoregressive encoder pass, training score regression together with a within-source ordering constraint over candidate summaries. On 80,856 human-reviewed Vietnamese document–summary instances, EViSE-mmBERT reaches an overall Spearman correlation of 0.806 on the held-out News test set and the compact EViSE-ViDeBERTa reaches 0.737; adding the ranking term to the same architecture raises the three criterion Spearman values by 0.045 to 0.061 over regression alone.

EViSE splits summary evaluation into faithfulness, coherence and relevance and returns all three scores from a single non-autoregressive encoder pass, training score regression together with a within-source ordering constraint over candidate summaries. On 80,856 human-reviewed Vietnamese document–summary instances, EViSE-mmBERT reaches an overall Spearman correlation of 0.806 on the held-out News test set and the compact EViSE-ViDeBERTa reaches 0.737; adding the ranking term to the same architecture raises the three criterion Spearman values by 0.045 to 0.061 over regression alone.

Cryo-EM and cryo-ET often produce density maps whose molecular identity is unknown, and at intermediate resolution the missing side-chain features make it hard to pick the right candidate from thousands of structures. CryoAssign combines CoLoRes rigid-body docking with learning-to-rank prioritization, searching intact density maps without prior segmentation and reranking candidates using several measures of map-model agreement. On 14 held-out benchmark maps it outperformed the original CoLoRes cross-correlation score and any single engineered feature; for the radial spoke complex it placed 11 of 18 known proteins in the top 15, where the original score placed only six.

Cryo-EM and cryo-ET often produce density maps whose molecular identity is unknown, and at intermediate resolution the missing side-chain features make it hard to pick the right candidate from thousands of structures. CryoAssign combines CoLoRes rigid-body docking with learning-to-rank prioritization, searching intact density maps without prior segmentation and reranking candidates using several measures of map-model agreement. On 14 held-out benchmark maps it outperformed the original CoLoRes cross-correlation score and any single engineered feature; for the radial spoke complex it placed 11 of 18 known proteins in the top 15, where the original score placed only six.