Research On Take-out Portrait Based On RFM And Extracted Features With XGBoost

With the rapid development of O2O(Online to Offline),the way of take-out has become a lifestyle which is heavily relied by people in the 21 st century,especially under the sudden outbreak of COVID-19 in recent years.Take-out brings more choices to people who stay at home.When new merchants entering the food delivery platform,consumers are the core marketing targets,and it is the most convenient and intuitive way to build accurate user portraits for them.In user usage scenarios,the matching of merchant portraits and consumption tendencies,as well as merchants’ marketing recommendations and discounts,are the keys to improving user stickiness and usage.With the improvement of big data technology and the iteration of various machine learning algorithms,more and more information can be mined from the static data as well as the user order transactions.This article attempts to construct portraits for merchants and users.Starting from the merchant’s point of view,the static label of the merchant is obtained through the crawler as a supplement,and the user’s order transactions are the main data source.Average unit price and order amount and average discount etc.are the main features to be selected,then we use KMeans cluster to classify the different types of merchants,which forms a part of the business’ s label.Secondly,from the user’s point of view,we can analyze the behavior data of the user’s order,and with the new feature of D which is the average order discount based on the traditional R,F,and M(Recency,Frequency,Monetary),serving as a complement for the traditional user segmentation.Several machine learning methods are used to construct the user turnover tendency prediction from the user order behavior data,and the XGBoost makes the best performance.The XGBoost is selected to figure out the features that have the highest impact on the user turnover tendency,then we use KMeans method to divide the users into three categories at last.For different types of users,with the analysis of their consumption characteristics,we also give the operational suggestions and strategies to improve their stickiness.Finally,for the second type of merchants with the great discounts and low average price,we get the POI data through GAODE API for supplementation.With the reasonable matching between merchant’s tags and user’s tags,we can connect them by the user portrait which can also be a complement for operational recommendations and strategies.This article summarizes the core and construction ideas of portraits.It innovatively adds the features of discount D to the basic R,F,M,as well as figuring out the features affects user turnover from order behavior data.And we classify users based on these important characteristics.At last,we associate the portraits of merchants and users which provides a data foundation for offering some operation strategies.