Research On Personalized Product Recommendation Method Based On Brand Community

With the rapid development of information technology and the wide application of mobile terminal devices,the Internet has entered the era of Web2.0,and e-commerce has rapidly developed into an indispensable commodity trading method in people's daily life.Human society has rapidly entered the era of information explosion.This creates an information overload problem.On the one hand,users cannot find the information that meets their needs from the vast amount of information;on the other hand,the information provider cannot provide effective information to the user for the personalized characteristics of the user.In the face of the predicament of information overload,the recommendation system becomes an effective means to solve this problem.With the stricter requirements of users for the accuracy of personalized information,the traditional recommendation model can no longer meet the needs of users,and the development of recommendation systems needs continuous improvement.Especially in the field of e-commerce,it is particularly important to improve the accuracy of product recommendation.At present,the main problems of system research are recommended: First,the cold start problem,including the user cold start,that is,the new user just entered the system,there is no historical transaction record related to it,and the system cannot provide accurate recommendation for the user.The cold start of the item,that is,the new product has not entered the market without the evaluation and purchase information related to it,and can not be recommended to the relevant users;the second is the data sparse problem,compared with the huge number of products sold by the website,the user has only scored the product.In the tip of the iceberg,this leads to the extremely sparse data of the user project scoring matrix.The accuracy of the user or the nearest neighbor of the project is relatively low,which makes the quality of the recommendation system drop drastically.In view of the above problems,this paper proposes a personalized product recommendation method based on brand community division.First,the user community is divided by brand recognition and user trust.Second,the coupledconvolutional neural network is used in the divided user groups.The model performs score prediction and recommends high-scoring products to users to improve the effect of product recommendation.This paper mainly improves the cold start and data sparseness of the article.The method first divides the user brand community,and uses the DBSCAN algorithm to cluster the user brand recognition and user activity obtained by analyzing the user purchase record and evaluation record.,initially formed a group.At the same time,through the user exchange information,the calculated user trust is matrixed with the preliminary group to form the final brand community;then the product score prediction is carried out in the user group,and the high-scoring product is recommended to the user.The scoring prediction model consists of two coupled parallel neural networks,a user network and a project network,divided into four layers(input layer,hidden layer,output layer,shared layer);user review data and commodity review data will be respectively User network and commodity network input,in terms of comment semantic analysis,will experiment from the perspective of word vector,and change the traditional mode of processing sentences using a single-sized convolution kernel,using multiple parallel convolution layers,using multiple Convolution kernels of different sizes are used to extract features;then the outputs of the two networks will be aggregated in the shared layer,and the machine learning FM algorithm is used for scoring prediction in the shared layer;finally,the high-scoring items in the group are recommended to the user,The accuracy of the product recommendation.This paper compares the mainstream recommendation methods including the classical collaborative filtering algorithm,and compares the accuracy,recall rate and F1 value.The experimental results show that the personalized product recommendation based on brand community division proposed in this paper The method has a better recommendation.