Tag-based self-learning task recommendation for mobile crowdsensing via collaborative multi-expert system

Abstract

With the powerful sensing capabilities of mobile smart devices, sensing users can complete mobile crowd sensing tasks through smart devices in the Internet of Things. When a user completes a sensing task, the sensing user may want to complete more related items from the previous task as an extended task. However, it is difficult for traditional recommender systems to satisfy sensing user needs for tasks because the recommendation results should be diverse. In this paper, we propose a multi-expert tag recommendation framework (MTRS) with the help of tags for sensing tasks. The main purpose is to predict whether the sensing user wants to extend the task, and provide appropriate related tasks as suggestions. The framework mainly consists of two parts: recommendation migration and recommendation purification. We extract features from multiple aspects. In the information gain expert network, information gain is obtained through the relationship between tags. In the feature interaction expert network, multi-head self-attention is used to model the feature interaction between different feature fields. In similar, the relevance between tags is explicitly emphasized in the degree expert network. Feature information is obtained through different expert networks, and a multi-critic strategy is proposed to judge the importance of each expert network. At the same time, different optimization goals are set for recommendation migration and recommendation purification. In recommendation purification, the user's satisfaction with task tags and the delay cost of task recommendation are specially considered. Finally, experiments demonstrate the correctness, effectiveness and robustness of the MTRS approach.