SAH: Shifting-Aware Asymmetric Hashing for Reverse k Maximum Inner Product Search

This paper investigates a new yet challenging problem called Reverse k-Maximum Inner Product Search (RkMIPS). Given query (item) vector, set of item vectors, and user the RkMIPS aims to find vectors whose inner products with vector are one k largest among vectors. We propose first subquadratic-time algorithm, i.e., Shifting-aware Asymmetric Hashing (SAH), tackle problem. To speed up Maximum (MIPS) on we design shifting-invariant asymmetric transformation develop novel sublinear-time Shifting-Aware Locality Sensitive (SA-ALSH) scheme. Furthermore, devise blocking strategy based Cone-Tree effectively prune (in batch). prove that SAH achieves theoretical guarantee for solving RMIPS Experimental results five real-world datasets show runs 4~8x faster than state-of-the-art methods while achieving F1-scores over 90%. The code is available at https://github.com/HuangQiang/SAH.