Research And Application Of Recommendation Algorithm Based On E-commerce Data

With the development of Internet technology,people have more and more ways to obtain information,the amount of information explodes,and the complexity of content also increases.Simple information retrieval technology has been unable to meet the needs of most people,and our era has been transformed from an era of scarcity to an era of abundance.With the increase in the number and variety of items,people's attention has become a scarce resource.E-commerce is indeed faced with the above problems,how to quickly find the items needed by users in the mass of data has become very important,so the personalized recommendation system under the scene of electric shopping mall came into being.At present,the rapid development of e-commerce,China is represented by taobao,jingdong,suning and so on,which contains a large number of users and commodity data,for a large amount of random data,the traditional recommendation algorithm can not well predict the pre-purchase of goods users.According to the above problem,I will recommend transformed to classification problems,use taobao user-commodity data set,the user behavior data feature extraction,using classification algorithm of machine learning,to predict user preferences of goods,the results found that the reinforcement learning GBDT model in F1 score than logistic regression algorithm of F1 score improved nearly doubled,GBDT algorithm in dealing with the randomness strong data has obvious advantages,the electricity recommend data sets obtained a good effect.Because under the electricity market scene according to the traditional recommendation algorithm using the score data sets,however goods data set faced with frequency points is lack or not according to the score data problems,aiming at this problem,this paper based on the traditional recommendation algorithm,the introduction of user behavior variables,on the basis of Jaccard similarity,put forward a new similarity calculation method between the user and user-goods score calculation method,through the combination of collaborative filtering recommendation between users.According to the taobao user behavior and item information data downloaded from alibaba's tianchi platform,some users were selected for the experiment,and the experimental result F1 score value was 0.154,and compared with the experimental result of traditional collaborative filtering algorithm,F1 score was greatly improved.Which can be concluded that the new user similarity computing method can more effectively find the target users of neighbor,so as to improve the accuracy of recommendation,and a new similarity calculation formula not used user rating data,so to a certain extent,solve the problem of the score data sparseness.