Task recommendation for mobile crowd sensing system based on multi-view user dynamic behavior prediction

Mobile crowd sensing is a data collection model that combines crowdsourcing ideas and mobile device sensing abilities. In the decision-making process of mobile crowd sensing perception behavior, a single type of historical behavior is used to predict the user's single preference tag, so the generalization ability of the model is weak, and the recommendation efficiency is not high. Aiming at the perception problem that the information overload of mobile crowd sensing leads to a significant increase in participants' decision-making costs, this paper proposes an innovative MUDBP prediction method based on Multi-view and social network group behavior to improve the task recommendation in mobile crowd sensing model. Specifically, this method starts from the multi-time behavior sequence, adopts an attention mechanism, sets different weights for different individual behaviors of various users according to the social influence of different users, and calculates the aggregation representation of group user behaviors at additional time granularity. Then, the multi-scale behavior sequence characteristics of a single user are fused with the multi-scale behavior sequence characteristics of a group of users in a social network, and the multi-view embedded behavior sequence characteristics of a single user are extracted. Finally, through multi-label prediction, the preference probability of users to various task types is obtained. Experimental results based on real data sets show that compared with other baseline methods, the proposed method can effectively improve the accuracy of task recommendation and reduce the perceived cost. At the same time, it can effectively deal with the cold start problem.