基于多源异构数据融合的移动群智感知任务推荐算法

Task Recommendation Based on Multi-Source Heterogeneous Data Fusion in Mobile Crowd Sensing

摘要: 现有的移动群智感知任务推荐方法主要包括基于时空信息和社交关系两大类，分别受到时空信息时变性和数据稀疏性问题的限制。为了突破上述限制并综合多源工人行为数据对任务推荐准确度的影响，本文提出了一种基于多源异构数据融合的移动群智感知任务推荐算法。该算法综合考虑了工人地理位置、时间可用性以及社交网络关系等多方面的因素，利用隐语义模型、双向长短时记忆网络模型分别生成空间、时间维度的工人–任务匹配度矩阵实现对工人偏好时空变化的挖掘，并利用Node2Vec图嵌入模型生成社交维度的工人–任务匹配度矩阵，缓解数据稀疏性问题。最后利用BP神经网络将上述三个矩阵进行融合以实现工人行为的多源异构数据融合，并最终进行任务推荐。实验结果表明，本文方法比单一数据源的任务推荐方法具有更好的推荐准确度。

Abstract: Existing task recommendation methods for mobile crowd sensing primarily fall into two categories: those based on spatial-temporal information and those based on social relationships, both constrained by the variability of spatial-temporal information and data sparsity issues. To overcome these limitations and enhance task recommendation accuracy by integrating multi-source worker behavior data, this paper proposes a task recommendation method based on multi-source heterogeneous data fusion for mobile crowd sensing. The proposed method comprehensively considers multiple factors, including workers’ geographic locations, time availability, and social network relationships. It employs a latent semantic model and a bi-directional long short-term memory model to generate worker-task matching matrices in the spatial and temporal dimensions to capture temporal-spatial variations in worker preferences, while the Node2Vec graph embedding model generates the worker-task matching matrix in the social dimension to alleviate data sparsity issues. Finally, a BP neural network is utilized to fuse these three matrices, achieving multi-source heterogeneous data fusion of worker behaviors and producing task recommendations. Experimental results demonstrate that the proposed method achieves better recommendation accuracy than task recommendation methods based on a single data source of worker behaviors.