Query-Aware Quantization for Maximum Inner Product Search

Maximum Inner Product Search (MIPS) plays an essential role in many applications ranging from information retrieval, recommender systems to natural language processing. However, exhaustive MIPS is often expensive and impractical when there are a large number of candidate items. The state-of-the-art quantization method approximated product with score-aware loss, developed by assuming that queries uniformly distributed the unit sphere. real-world datasets, above assumption about does not necessarily hold. To this end, we propose based on distribution combined sampled softmax. Further, introduce general framework encompassing proposed multiple methods, develop effective optimization for framework. evaluated three datasets. experimental results show it outperforms baselines.