Running summarizing tools on a new article

Abstract: The Hierarchical Navigable Small World (HNSW) index has become a mainstream technology in the field of approximate nearest neighbor (ANN) search due to its excellent query efficiency and high recall rate. However, the prohibitively long construction time of the HNSW index is a significant performance bottleneck, especially when handling large-scale datasets. The core overhead during the construction process lies in the complex search for multilayer neighbors of newly inserted points. To address this issue, this paper proposes an HNSW index construction optimization method based on Hierarchical Distance-Max constraint. This method introduces a configurable parameter$A$during the neighbor search process and crucially, dynamically adjusts the constraint strength according to the index hierarchy. By limiting the maximum distance for neighbor searches and pruning distant potential neighbors early in the search phase, the H-DMax mechanism significantly reduces the computational complexity of each insertion operation while, through fine control of A, strictly preserving the long-range connectivity at higher layers. Experimental results (to be added) show that the H-DMax method can improve HNSW construction speed by up to 1.27x while maintaining high recall, providing a more efficient solution for large-scale ANN applications.