1(College of Information,Liaoning University,Shenyang 110036,China)2(Computer Department,Liaoning Vocational College of Light Industry,Dalian 116100,China)3(North China Chemical Sales Branch,Petro China Co Ltd.,Zhengzhou 450000,China)
Abstract:A large number of comment data in e-commerce contain abundant information,the information helps to solve the problem of data sparsity in personalized recommendation system.In order to improve the efficiency of using comment data and the accuracy of commodity recommendation,a personalized commodity recommendation method based on coupled CNN scoring predictive model was proposed.This method uses CNN to construct scoring prediction model and divides the coupled CNN into user network and commodity network,which are divided into input layer,convolution layer,output layer and sharing layer.User comment and commodity comment are input from corresponding network respectively.In the analysis of commentary data,semantic analysis is carried out from the perspective of word vector,while changing the traditional sentence processing mode using single-size convolution kernel,using multiple parallel convolution layers,using multiple convolution kernels with different sizes to extract sentence features.The outputs of the two networks are converged in the sharing layer,where the Factorization Machine algorithm is used for scoring prediction.Finally,the high-score commodities in the results are recommended to users.The results of comparative experiments show that the proposed method can improve the accuracy of commodity recommendation.