一种融合信息网络结构的数据增强行为预测算法

Data Enhanced Behavior Prediction Algorithm Based on Information Network Structure

随着互联网上用户移动数据的日益繁荣,用户的移动行为预测也成为了预测研究的热点.近年来,循环神经网络(RNN)技术因其高效性和扩展性在移动预测中得到了广泛的应用.但是,目前大部分网上收集到的用户移动行为数据普遍具有稀疏和异质的特性,特别是当用户出于习惯或隐私考虑可能会拒绝向平台提交活动记录.因此在这些稀疏数据集上基于RNN的预测技术无法有效地学习到足够的用户行为特征,从而影响了模型的预测性能.为了解决该问题,本文提出了一种融合信息网络结构的数据增强行为预测算法.具体来说,首先我们将用户历史行为数据转为信息网络图;然后通过该信息网络的模块度来评估用户的信息传递效率;最后根据信息传递效率对用户的朋友数据进行采样,将具有高信息传递效率的朋友数据嵌入到用户数据中对用户数据进行增强.在真实数据集Yelp上的实验结果显示,我们的方法可以起到对现有算法模型增强的作用,所有模型的预测性能都得到了大幅提升.

With the rapid increase of user mobile data on the Internet,the prediction of user′s mobility behavior has become a hot spot for prediction research.In recent years,the recurrent neural network(RNN)technology has been widely used in mobile prediction because of its high efficiency and scalability.However,most of the mobile behavior data collected from the Internet are sparse and heterogeneous,especially when users may refuse to submit activity records to the platform for privacy reasons.Therefore,RNN based prediction technology can not effectively learn enough user behavior features on these sparse data sets,which affects the prediction performance of the model.To solve this problem,this paper proposes a data-enhanced behavior prediction algorithm based on information network structure.Specifically,firstly,we transform the user′s historical behavior data into an information network graph;secondly,we evaluate the information transmission efficiency of users through the modularity of the information network;finally,we sample the user′s friend data according to the information transmission efficiency,and embed the friend data with high information transmission efficiency into the user data to enhance the user data.The experimental results on the real yelp dataset show that our method can enhance the existing algorithm models,and the prediction performance of all models has been greatly improved.