AUC-Optimized Training Improves Neural Retrieval Accuracy

Researchers propose MW loss, directly maximizing the Mann‑Whitney U statistic and theoretically upper‑bounding AUC, achieving higher AUC and recall on retrieval benchmarks. Read more: getnews.me/auc-optimized-training-i... #dualencoder #retrieval

AIRwaves Boosts Tweet-to-Study Retrieval with Dual Encoders

AIRwaves’s two‑stage system raised MRR@5 to 0.6828, up from the BM25 baseline of 0.5025, using an E5‑large encoder and SciBERT re‑ranker on biomedical and computational‑science tweets. Read more: getnews.me/airwaves-boosts-tweet-to... #airwaves #dualencoder

Hierarchical Retrieval Advances with Pretrain‑Finetune Method

A two-stage pretrain-finetune recipe lifts hierarchical retrieval recall from 19% to 76% for distant ancestors, while keeping close-match performance, per a Sep 2025 study. getnews.me/hierarchical-retrieval-a... #hierarchicalretrieval #dualencoder