Research On Brand Recommendation Of E-commerce Website Based On User Time-evolving Behavior Features

In e-commerce,searching is hard to meet users' daily needs,users need new ways to get their favorite items.E-commerce recommendation system can help users find their needs in the mass of items and build a bridge that directly communicates users and sellers.The construction of e-commerce recommendation system needs to make a full analysis of the data and find out the useful features of the recommendation and ensure the effectiveness of the system.However,the current analysis only focuses on the simple behavior changes and does not pay attention to the interaction and evolution of different behaviors with time.Another problem of e-commerce recommendation system is that with the influx of new users and new items,the sparsity of data is relatively large.It is hard to find similar users when constructing a collaborative filtering system based on the sparsity matrix.In this paper,the recommendation system of e-commerce is studied,and the following work is done according to the above problems.1)In this paper,a real e-business data set is analyzed.By taking the user's brand purchasing intention as the analysis center,the evolution and mutual influence of different behavior in brand purchase are found through the fine-grained analysis,as well as the attributes of different users when they purchase the brand.There are three kinds of behaviors of users on e-commerce platform,including the click behavior,the collection behavior and the purchase behavior.Each behavior can reflect a certain user purchase intention.At the same time,the user's attribute data can also affect the user's brand purchase.These analyses can help to better understand users' buying tendency.2)A brand purchase prediction based on time-evolving features is proposed.In this paper,the user behavior data and the user attribute data are extracted,and the features of user timeevolving,user attribute and brand attribute are obtained.And through the promotion activities to predict and predict the daily purchase of the two scenarios,the effectiveness of time-evolving features is verified.At the same time,the different importance of different features in the two scenarios reflects the difference of users' psychology.Finally,a brand recommendation algorithm is designed.Brand recommendation is based on collaborative filtering algorithm.From the perspective of user similarity,the similarity of two users in multiple dimensions is calculated by multi-dimensional features.Two different sets of data are selected to train two models according to the selected classifier.The similarity prediction is combined with the results of two models.In the experiment,a certain number of users are randomly selected to verify that the algorithm novelty and compared with the traditional collaborative filtering,it improves the accuracy.