With the Internet of Things (IoT) becoming the infrastructure to support domain applications, IoT search engines have attracted increasing attention from users, industry, and research community, since they are capable of crawling heterogeneous data sources in a highly dynamic environment. IoT search engines have to be able to process tens of thousands of spatial-time-keyword queries per second, making query throughput a critical issue. To achieve this heavy workload, caching mechanisms in collaborative edge-cloud computing architecture, which can implement the caching paradigm in cloud for frequent n -hop neighbor activity regions, is first proposed in this article. With our design, the frequent query result can be gained quickly from the spatial-time-keyword filtering index of n -hop neighbor regions by modeling keywords relevance and uncertain traveling time. In addition, we use STK-tree proposed previously to directly answer nonfrequent queries. Extensive experiments on real-life and synthetic data sets demonstrate that our proposed method outperforms the state-of-the-art approaches with respect to query time and message number.