OPERA: Online Data Pruning for Efficient Retrieval Model Adaptation

Collection of recent retrieval and recommendation research (arXiv, 19 Mar 2026) that advances efficiency, robustness, and deployability of dense and generative retrieval systems. Key contributions: OPERA (static and two-stage dynamic data pruning) to accelerate and improve dense retriever fine-tuning (DP reaches comparable performance in <50% training time); FCUCR, a federated continual recommendation framework using time-aware self-distillation and inter-user prototype transfer for long-term personalization; Profiler + DAVINCI for citation recommendation with a proposed inductive evaluation;P