Knowledge-Geometry Decoupling: Refreshable Pretrained Transfer for Streaming Recommendation

The document is an arXiv computer-science information-retrieval feed containing new research on recommender systems, deep-research agents, retrieval and reranking, multi-agent collaborative-filtering security, bandit recommendation, OOD recommendation, and long-sequence modeling. One paper specifically studies attacks and defenses against multi-agent collaborative-filtering systems, including dissemination and extraction attacks whose effectiveness varies with system connectivity. The remaining papers primarily present performance, compression, retrieval-quality, robustness, or deployment-e 관련