Evaluating spatial-keyword queries on streaming data

Abstract

This paper provides an extensive experimental evaluation for different spatial-keyword index structures on streaming data. We extend existing snapshot spatial-keyword queries with the temporal dimension to effectively serve streaming data applications. Then, the major index structures are equipped with efficient query processing techniques and evaluated to process the extended queries. The evaluation is oriented towards a system building perspective to provide system builders with insights on supporting scalable spatial-keyword queries on fast data streams, e.g., social media streams and news streams. In particular, we have taken existing spatial-keyword index structures apart into four major building blocks that are commonly supported at a system-level. Ten different index structures are then composed as combinations of these four building blocks. The ten indexes are wholly residents in main-memory, and they are evaluated on real datasets and query locations. The index performance is measured in terms of data digestion rate in real time, main-memory footprint, and query latency. The results show the relative performance gains of both basic and hybrid index structures with abundant insights from a system point of view.