Research On Model And Algorithm Of Personalized Recommendation System Based On User Shopping Information

With the continuous development of the current Internet technology,more and more businesses are now developing their own online e-commerce through the Internet.With the arrival of Web 2.0 and "Internet +",the Internet has become indispensable in people's daily life.Therefore,when we use the Internet,huge amounts of data and information are presented to us.With the continuous popularization of the Internet,this problem will become more prominent,it will make the consumers find it difficult to meet their interest things,businesses also face the risk of user loss.In order to solve the problem of information overload and information explosion,the personalized recommendation system came into being.Personalized recommendation system to calculate the similarity between users by mining the user's historical information or the neighbor information to achieve the recommended function for the user.At present,there are three main problems to be considered in the system of personalized recommendation: Firstly,with the growth of users and projects,how to avoid the sparsity malpractice of the recommended system.Secondly,how to efficiently and quickly get those complete and reliable user information from the mass data.Thirdly,how to recommend related things according to time information in the recommendation process.In order to provide better personalized recommendation for users and businesses,the paper will study the above three questions.Firstly,the paper proposes a collaborative filtering algorithm based on balance parameters.The algorithm mainly aims at the sparsity problem of traditional collaborative filtering,in the traditional algorithm,user scoring matrices often generate null values,so that users with different interests have the same similarity in calculating user similarity,The paper proposes the balance parameter to solve this problem.The balance parameter is calculated according to the user's attribute difference.Each data set is different and the result is different.Then the balance parameter is combined with the traditional cosine similarity calculation method to achieve the effect of balance user similarity.From the experimental results,this method alleviates the sparsity problem of the traditional collaborative filtering algorithm.Secondly,the paper builds the data preprocessing framework and carries out data processing.Due to the redundant and missing data used in the experiment,we use the Hadoop cluster to filter the data and built a three-node Hadoop data processing cluster,and then use MapReduce program to clean up the data set so as to get reliable user information.Finally,a recommendation algorithm based on time awareness label is proposed in this paper.This algorithm mainly aims at the problem of insensitive recommendation time,in the process of calculation of the weights of the user,according to the time the user purchased the item,we use a new weight calculation method,we use self applicable functions to alleviate the insensitivity of the recommended algorithm in time.At the same time,the improved collaborative filtering algorithm is combined with the time awareness label model,and the specific steps of the algorithm are proposed in this paper.Through the analysis of the experimental results,accuracy,recall and other performance indicators have improved to a certain extent,and the paper also alleviates the problem of insensitive recommendation time.Overall,the paper improves on the three problems existing in the personalized recommendation system.From the experimental results,the improved recommendation algorithm has a higher performance and accuracy.