RecBundle: A Next-Generation Geometric Paradigm for Explainable Recommender Systems

ArXi:2603.16088v1 Announce Type: cross Recommender systems are inherently dynamic feedback loops where prolonged local interactions accumulate into macroscopic structural degradation such as information cocoons. Existing representation learning paradigms are universally constrained by the assumption of a single flat space, forcing topologically grounded user associations and semantically driven historical interactions to be fitted within the same vector space.