Prediction Of Click-to-buy Of E-commerce Users Based On Graph Convolution Neural Network

E-commerce has just risen to become an important role in people's daily life after only three or four years,China's e-commerce technology industry has formed a very large and complete ecosystem.The rapid development of e-commerce is closely related to its convenient,fast,time-saving and labor-saving features.At the same time,because there is no need for face-to-face transactions between businesses and users in e-commerce,businesses cannot accurately grasp the preferences and needs of users.With the rapid development of large-scale data industry,e-commerce industry is also rising rapidly.In this industry,after the transaction between users and the platform,the platform will leave a lot of relevant data,including user characteristic data,user behavior data and product characteristic data that users take.If we can effectively use these data to accurately analyze and predict the user's behavior,we can recommend their preferred products for users,and then improve the user's browsing,clicking and even purchasing volume of each recommended product.This article takes the real user characteristic data,commodity characteristic data and user behavior data left by users in the e-commerce platform Alibaba as the original data,and models the prediction of users' click to buy behavior through the method of graph neural network.Firstly,this paper analyzes the research background and significance of predicting user's click to buy behavior in e-commerce.If we can accurately predict the user's click preference,it is of great significance to improve the business volume of e-commerce platform,meet the user's needs,stimulate the user's desire to buy and save time.After that,the paper reviews the literature of the prediction and research of user's click behavior,which helps understand the current situation of the research and its shortcomings.Next,the paper analyzes original data,by mainly discussing the relationship among various characteristics of users and user behaviors.Finally,it is the modeling steps,including preprocessing data,building the prediction model and the comparative analysis of the model effect.In order to integrate user's behavior data and recommend preferred products to users,this paper proposes a data analysis method graph neural network to predict user's click to buy behavior.Firstly,the user commodity behavior data is processed into a bipartite graph matrix.At the same time,the user commodity frequency filtering method is used to eliminate the invalid behavior of the user behavior data,and the positive and negative samples are generated by randomly adding the commodity behavior to the user.Then,the bipartite graph matrix is convoluted.After the convolution,the user itself is used as the root node to fit the transfer function of single-layer neural network,and the information of other nodes is transferred to the root node.After the convolution of bipartite graph matrix,the user itself is used as the root node to fit the transfer function of single-layer neural network,the information of other nodes is transferred to the root node,and then the user click behavior in the graph is normal according to the information of the root node.Finally,the available prediction model of user click-purchase behavior is obtained.After developing the prediction model,we need to evaluate its effectiveness.Because the single evaluation index cannot fully explain the advantages and disadvantages of the prediction model,this paper also introduces two algorithms that are commonly used in user recommendation systems:collaborative filtering and DeepFM.The results of the three methods are compared and analyzed,and it shows the results of the prediction model based on the graph neural network and the prediction model based on DeepFM are relatively stable.However,the results of the prediction model based on collaborative filtering fluctuate greatly and the stability of the model is insufficient.In addition,this paper also compares the running time and the maximum memory consumption of the three models.The model based on collaborative filtering takes the least time and the minimum memory consumption.The prediction model based on graph neural network takes the longest time and takes the most memory.However,the overall performance of the model based on graph neural network is the best.The research of this paper is based on real data in the e-commerce scene,aiming to build a prediction model that can be used to predict user behaviors.Although the current research is still in the theoretical stage,in the future,the prediction model obtained in this paper is likely to be directly applied to the recommendation system of real e-commerce websites.