Matching Subscription Over Geo-Textual Streams from IoT via Social-Aware Clustering and Apache Flink

Abstract

Current location-based services (LBS) continuously generate a massive amount of geo-message streams. The cluster-based subscription matching method is an effective means to feed subscribers with related geo-messages from geo-message streaming. However, current cluster-based subscription matching methods only consider the spatial relationship and textual relationship and ignore users’ social relationship. As a result, the matching results may not completely satisfy the requirements of users. In this paper, we proposed a social-aware subscription matching method by taking spatial, textual, and social factors into consideration. Then, we used a cache strategy and a Flink-based acceleration process to reduce the extra time overhead caused by computing the social relationships. A set of extensive experiments have been conducted on a real dataset. The experimental results indicate that our method improves the recall of matching results. Besides, the Flink-based acceleration process with caching can speed up the subscription matching process by a ratio of up to 3.299 compared with the state-of-the-art.