基于ResNet和DF融合的用户购买预测算法研究

Research on User Purchase Prediction Algorithm Based on the Fusion of ResNet and DF

摘要: 各大电商平台在前期为客户提供服务的同时已经积累了大量用户及商品数据，如何充分利用这些数据为企业增加收入、为用户提供个性化服务已成为研究热点。基于电商平台的环境及数据情况，针对电商平台用户及商品种类数量众多，但平台方无法准确预测用户是否购买这一问题，本文提出了一种基于残差神经网络(ResNet)和深度森林(Deep Forest)融合的用户购买行为预测算法。对某在线商城的大量数据处理为150维的用户特征数据和120维的商品特征数据。首先利用残差神经网络对用户购买行为进行预测，后通过深度森林进行预测，最后通过线性叠加的方式将两种模型融合。通过对残差神经网络进行调参，对深度森林中的随机森林深度进行调整进一步提高预测精度。实验结果表明，该融合模型相比传统算法具有更好的预测效果。

Abstract: Major e-commerce platforms have accumulated a large amount of user and product data while providing services to customers in the early stage. How to make full use of these data to increase revenue for enterprises and provide personalized services for users has become a research hotspot. Based on the environment and data of the e-commerce platform, in view of the large number of e-commerce platform users and product types, the platform which cannot accurately predict whether the user will buy or not, this paper proposes a user purchase behavior prediction algorithm combined the residual neural network (ResNet) with deep forest (Deep Forest). A large amount of data of an online shopping mall is processed into 150-dimensional user characteristic data and 120-dimensional commodity characteristic data. First, the residual neural network is used to predict the user’s purchase behavior, and then the deep forest is used to predict, and finally the two models are merged by linear superposition. By adjusting the parameters of the residual neural network, the depth of the random forest in the deep forest is adjusted to further improve the prediction accuracy. Experimental results show that the fusion model has a better prediction effect than traditional algorithms.