An Efficient Indexing Approach for Continuous Spatial Approximate Keyword Queries over Geo-Textual Streaming Data

Current social-network-based and location-based-service applications need to handle continuous spatial approximate keyword queries over geo-textual streaming data of high density. The continuous query is a well-known expensive operation. The optimization of continuous query processing is still an open issue. For geo-textual streaming data, the performance issue is more serious since both location information and textual description need to be matched for each incoming streaming data tuple. The state-of-the-art continuous spatial-keyword query indexing approaches generally lack both support for approximate keyword matching and high-performance processing for geo-textual streaming data. Aiming to tackle this problem, this paper first proposes an indexing approach for efficient supporting of continuous spatial approximate keyword queries by integrating $min-wise$ signatures into an AP-tree, namely AP-tree ${}^{+}$ . AP-tree ${}^{+}$ utilizes the one-permutation $min-wise$ hashing method to achieve a much lower signature maintenance costs compared with the traditional $min-wise$ hashing method because it only employs one hashing function instead of dozens. Towards providing a more efficient indexing approach, this paper has explored the feasibility of parallelizing AP-tree ${}^{+}$ by employing a Graphic Processing Unit (GPU). We mapped the AP-tree ${}^{+}$ data structure into the GPU’s memory with a variety of one-dimensional arrays to form the GPU-aided AP-tree ${}^{+}$ . Furthermore, a $min-wise$ parallel hashing algorithm with a scheme of data parallel and a GPU-CPU data communication method based on a four-stage pipeline way have been used to optimize the performance of the GPU-aided AP-tree ${}^{+}$ . The experimental results indicate that (1) AP-tree ${}^{+}$ can reduce the space cost by about 11% compared with MHR-tree, (2) AP-tree ${}^{+}$ can hold a comparable recall and 5.64× query performance gain compared with MHR-tree while saving 41.66% maintenance cost on average, (3) the GPU-aided AP-tree ${}^{+}$ can attain an average speedup of 5.76× compared to AP-tree ${}^{+}$ , and (4) the GPU-CPU data communication scheme can further improve the query performance of the GPU-aided AP-tree ${}^{+}$ by 39.4%.