Brand Preference Prediction Method Integrating User Social Information

The Internet has promoted the explosive growth of information,which makes it increasingly difficult for people to locate information of interest from massive information.The problem is known as "information overload".To effectively solve this problem,the personalized recommendation technology obtains users' preference by mining massive data,on which the recommendation of the corresponding content is based.However,personalized recommendation technology still faces challenges such as data sparsity and cold-start problem.Social sites provide an additional source of information for mining users' interests.Many studies address the issues faced by traditional personalized recommendations by introducing users' social information across domains.However,most studies only consider the introduction of users' personal information and social relationship information,and rarely consider their social behavior information.At the same time,there are three challenges when introducing social behavior information across domains: 1)The social domain and e-commerce domain data usually come from different websites.Hence,introducing social information needs to find overlapping users between domains,and aligning information in different fields through these users.2)Data in different fields are in different forms.Features built from different fields must be input into the same recommendation model as well as maintain the information integrity of the original data.3)The time when social behavior and shopping behavior occur is not synchronized so the introduction of time series information needs to avoid the interference caused by behavioral none synchronization.In order to cope with these challenges,this paper proposes a cross-domain temporal preference mining algorithm,which effectively introduces social behavior information and alleviates the problems of users' cold-start and changes of users' interest.The main contributions are as follows:1.This paper propose a cross-domain temporal preference mining algorithm based on matrix decomposition.Firstly,based on matrix decomposition technique,a cross-domain preference prediction model is proposed to construct a cross-domain mapping relationship between social behavior characteristics and commodity purchase preferences.Then,from the time dimension,it is assumed that users have different interests in different time periods,and the behaviors are divided into different time periods according to the time when the social and shopping behaviors occur.The corresponding time series features are constructed.Finally,a cross-domain temporal preference prediction model is proposed to mine users' sequential purchase preferences from their temporal social behavior,recommend the products that users are interested in,and solve problems such as user cold start and user interest changes.2.This paper obtains and constructs the Weibo data through its open source API,and the purchase data from a large domestic e-commerce platform.By overlapping users to connect data from different fields,the cross-domain data is constructed.Experiments on this data demonstrate the effectiveness of our proposed model.3.This paper further proposes a construction scheme of a potential customer mining and recommendation system,which is based on the proposed cross-domain temporal preference prediction method and can be applied in the online environment.This system scheme can effectively solve the problem of preference prediction and recommendation from the new users logged in through the social account according to their social information.Furthermore,it can timely update the prediction results based on the changes of users' social information,thereby providing guarantees for the recommendation effects in the system.In conclusion,our paper verifies the assumption that our method can efficiently improve the recommendation effect with users' social text behavior and the way of cross-domain.It also verifies that the proposed cross-domain temporal preference mining algorithm can better solve the problems of user cold start and user interest change,and effectively improve the recommendation quality of the system.