A learned spatial textual index for efficient keyword queries

Spatial textual indexing techniques allow one to efficiently access and process large volume of geospatial data, and recent research efforts have demonstrated that learned indexes can lead to better performance in comparison to conventional indexes. In this paper, we present a learned spatial textual index designed to process spatial textual data efficiently. Specifically, our proposed index is constructed based on the idea of radix table, spline points, and inverted lists. Besides, Morton encoding was used to convert high-dimensional coordinates into one dimension. In order to handle data insertion, deletion, and update in real-time, a gap array is used to store the underlying data, and a space reallocation strategy in units of spline points is designed. Based on the index, we propose query processing algorithms to handle different spatial keyword queries efficiently. An optimizer using random forest regression model was also designed to obtain appropriate index parameters for minimizing query latency. We evaluate our proposed index with IR-tree, and the findings show that our index outperforms IR-tree in terms of construction time, index size, and query efficiency.

Automated summary

This paper presents a learned spatial textual index for efficiently processing spatial textual data. The index is constructed based on radix table, spline points, and inverted lists, with high-dimensional coordinates converted using Morton encoding. Real-time data insertion, deletion, and update are handled using a gap array and a space reallocation strategy. Query processing algorithms and an optimizer using random forest regression model are proposed to enhance query efficiency. Evaluation results show that the proposed index outperforms the IR-tree in terms of construction time, index size, and query efficiency.