Task recommendation method for fusion of multi-view social relationship learning and reasoning in the mobile crowd sensing system

Abstract

With the powerful sensing ability of mobile smart devices, sensing users can obtain crowd sensing services for intelligent devices in the Internet of Things. However, due to the low accuracy of task recommendations, the uneven quality of data provided by sensing users has always been a problem. In this paper, a novel mobile crowd sensing task recommendation framework is proposed by integrating multi-view social relationship reasoning, which aims to construct a social relation network through multi-view to ensure task recommendation accuracy and improve data quality. Firstly, the social relationship network is jointly built by location matching view, time series view, and preference view. Secondly, the sensing user preferences and candidate task representations are learned by extracting task name features, task subject features, and task category features. Finally, the key nodes are used to divide the social relationship sub-network. The preference representation of the key nodes and the candidate task representation is used to do the inner product to obtain the probability of the social relationship sub-network selection task, and then complete the task recommendation. Evaluations based on two real datasets, Gowalla and Brightkite, show that the average recommendation accuracy is about 93%, and the participation rate of sensing users is about 97%. At the same time, the running time is reduced by 15% on average compared with the baseline algorithm, and the mobile cost of sensing users is reduced.