基于少样本学习和图神经网络的移动群智感知招募方法

A Recruitment Method for Mobile Crowd Sensing Based on Few-Shot Learning and Graph Neural Networks

摘要: 在移动群智感知招募中，当工人未执行任务或执行任务较少时，工人的偏好特征和信誉度特征信息存在稀疏性问题，考虑工人偏好和信誉度特征的招募方法很难准确估计工人的任务完成质量。同时现有研究大多忽略了工人待完成任务对工人招募的客观影响。为此本文提出了一种基于少样本学习和图神经网络的移动群智感知招募方法。首先，建立基于图神经网络的小样本学习模型(FSL-GNN)的工人与任务主观相关性估计模型，估计出工人的工人偏好对任务特征的相关性和工人信誉度对任务特征的相关性。然后，设计基于异质图卷积网络的工人与任务的客观相关性模型，计算工人待完成任务与待招募任务特征之间的客观相关性。最后使用注意力机制结合以上三种相关性估计出工人对任务的完成质量。基于大量真实数据集的实验仿真结果表明，提出的招募方法具有较好的工人任务完成质量估计准确度优势，与其他招募模型相比显著提高了任务的完成质量。

Abstract: In existing mobile crowd sensing recruitment methods, data sparsity issues exist when there are workers who have few or no experience of executing tasks, which limits the analysis of preference and credibility features while estimating the task completion quality of workers. In addition, these studies ignore the objective impact of incomplete tasks in the hiring process. At the same time, most of the existing studies ignore the objective impact of the tasks to be completed by workers on worker recruitment. To overcome those problems, this paper proposes a novel recruitment method for mobile crowd sensing based on few-shot learning and graph neural networks (FSL-GNN). First, a worker-task subjective correlation estimation model is built on a few-shot learning framework with a graph neural network, which is employed to estimate two types of subjective correlations: worker preferences to task features and worker credibility related to task features. Then, this study designs an objective correlation model between workers and tasks based on the convolutional network of heterogeneous graphs, and calculates the objective correlation between the characteristics of workers’ tasks to be completed and tasks to be recruited. Finally, the attention mechanism combined with the above three correlations is used to estimate the workers’ completion quality of the task. Experimental simulations on extensive real-world datasets demonstrate that the proposed method can provide accurate task completion quality estimation and outperforms other recruitment models in terms of task quality enhancement.