LIRA: A Learning-based Query-aware Partition Framework for Large-scale ANN Search

Approximate nearest neighbor search is fundamental in information retrieval. Previous partition-based methods enhance search efficiency by probing partial partitions, yet they face two common issues. In the query phase, a common strategy is to probe partitions based on the distance ranks of a query to partition centroids, which inevitably probes irrelevant partitions as it ignores data distribution. In the partition construction phase, all partition-based methods face the boundary problem that separates a query's nearest neighbors to multiple partitions, resulting in a long-tailed 𝑘NN distribution and degrading the optimal 𝑛𝑝𝑟𝑜𝑏𝑒 (i.e., the number of probing partitions). To address this gap, we propose LIRA, a LearnIng-based queRy-aware pArtition framework. Specifically, we propose a probing model to directly probe the partitions containing the 𝑘NN of a query, which can reduce probing waste and allow for queryaware probing with 𝑛𝑝𝑟𝑜𝑏𝑒 individually. Moreover, we incorporate the probing model into a learning-based redundancy strategy to mitigate the adverse impact of the long-tailed 𝑘NN distribution on search efficiency. Extensive experiments on real-world vector datasets demonstrate the superiority of LIRA in the trade-off among accuracy, latency, and query fan-out. The codes are available at https://github.com/SimoneZeng/LIRA-ANN-search . CCS Concepts • Information systems → Information retrieval query processing; Learning to rank; Top-k retrieval in databases.