Customer Value Research And Based On Improved RFM Model Recommendation Prediction

In the online shopping scene,the virtual display of goods breaks the space restriction of traditional shelves and provides consumers with massive information and goods,but also causes the problem of excessive choice and information overload,which is not only bad for merchants’ profits,but also damages consumer satisfaction and loyalty.Therefore,ecommerce platforms or groups hope to push products to high-value groups and important consumers,while consumers hope to obtain products that meet their individual needs from a large number of products.Firstly,the product order data is used to analyze the consumer targeted marketing strategy and establish the RFM model in the marketing field.According to the level of R,F and M,eight types of customers are distinguished.After the keyword grouping of product description and product price is extracted,it is found that price division has a great impact on classification,so a new index C is introduced: the average price of different commodities.Secondly,in order to avoid the impact of different business scenario requirements,the weight coefficient is added to the RFMC model.After classification,it is found that the weighted RFMC model stratification has certain effects,and there are obvious differences among various customers.Then attribute specification and data transformation of the four indicators,using the K-Means algorithm for customer clustering,observation SSE and the number of cluster relationship,using the elbow method to determine the parameter 5 is the number of clustering,so as to analyze the characteristics of customer group.In the aspect of product recommendation system design,through the average calculation of each customer and recommended products and prices,it is concluded that in the product catalog of more than 3000 products,the FP Growth model based on association rules can predict the next product that customers will buy in 35% of the cases,thus generating significant additional income.In the collaborative filtering recommendation algorithm based on multi-type similarity calculation,comparative experiments are added to multiple similarity calculation methods.It is found that for collaborative filtering based on user and commodity,RMSE calculated by European distance is the smallest,which is 5.6884 and 5.7720 respectively.After that,customers are randomly selected and substituted into the collaborative filtering algorithm based on Euclidean distance similarity measurement to obtain different commodity-user recommendation results.Previous studies aimed at improving the rationality of new variables in the RFM model often have not been proved relevant.In order to change this defect,before improving the new model,text analysis is adopted to dig out a reasonable new variable,namely the average price of different commodities,and corresponding supporting conclusions are obtained in product and customer classification.