Research On User Portrait Based On Machine Learning

In recent years,with the development of China's intelligent technology,various kinds of apps have penetrated into our daily life,and the recorded information has an explosive growth trend.Massive data has started the transformation of the times.Enterprises in various fields have begun to pay attention to data and study how to effectively use the massive data to extract valuable things.Tiktok's products like content recommendation and E-commerce guess you like algorithm are created.Online retailers Operators are to use user behavior data to build user profiles,so as to realize refined operation for enterprises.For businesses,seizing the opportunity is equal to seizing the user's wallet.However,the traditional method of collecting data to build the target user group is inefficient.In order to effectively achieve the accurate marketing of major businesses,researchers have launched the method of studying user portrait based on algorithm model,which has become the mainstream.In the feature extraction stage,we use the time sliding window method to extract the user's behavior characteristics in the past month and three days,and then collect the ratio of the user's behavior in one month minus,the behavior in the last three days to the behavior in that month.In this way,the more distant the user's interaction date is from the forecast date,the smaller the weight index of the model forecast.In the first mock exam,we will study the concept and construction method of user portrait based on the behavior data of real users in Jingdong online shopping mall,and predict users' purchase intention.First,this article sets up the first mock exam.The first mock exam is compared with the three single models of Xgboost,GBDT and Lightbgm,the optimal single model is obtained.Then the combination weighting models are compared to select the optimal combination model to predict the intention users in recent days,so as to provide accurate target groups for consumers.After verification,this paper concludes that xgboost-gbdt-lightgbm combination model can better predict the purchase intention customers,and the accuracy rate,recall rate and F1 of its model prediction are 0.873?0.706?0.834.Using this method for reference,it can be applied to improve the precision marketing ability of e-commerce stores,put advertising intention customers,and reduce operating costs.