A Study On Predicting The Repeating Purchase Behavior Of Community E-commerce Customers And Recommendation Algorithms

As the e-commerce develops rapidly,more and more people begin to make their purchases through community-based e-commerce platforms.Those purchases provide abundant operating data for e-commerce companies.Among those data,users' preferences are reflected by their repeating purchase behaviors which is one of the main profit sources for a company.The community-based e-commerce platform needs to figure out how to predict their clients' repeating purchase behaviors based on big data and enhance their customers' willingness to conduct such repetition.Repeating purchase behavior prediction can be used in the personalized recommendation system to identify customers with repeated purchase intention,so as to achieve the purpose of precision marketing and provide personalized services for users.The key process of predicting users' repeating purchases is to identify the user's behavior pattern hiding in the data through proper Model algorithms.Some scholars have already carried out researches on the e-commerce platforms recommendation algorithms based on data statistics law,but there are still many unsolved issues in the practical application for researchers to dig deeper.Among which,how to establish proper recommendation algorithms based on e-commerce platforms' data and how to predict users' repeating purchase behaviors have become two of the most popular research interests for both the academia world and e-commerce companies.This research proposes a way to predict users' repeating purchase behaviors and a recommendation algorithm based on a community-based ecommerce platform's users' characteristics through mathematical modeling and the practical operating data.Firstly,the dissertation proposes an ameliorated RFM model to extract the user's characteristics and analyzes the user group by clusters.The improved model could handle the two distinctive characteristics of community-based e-commerce platforms better,which are the stability of the user group and their repeating consumption pattern on certain items.Secondly,the research focuses on the prediction of the customer's purchase behaviors based on the improved RFM model which profiles the customers' purchase behavior patterns.The prediction is accomplished by analyzing the data distribution of the customer's purchase behaviors.To guarantee the accuracy of the community-based platform's recommendation on the commodity,the study analyzes the characteristics of T-APP's marketing data and then proposes a recommendation algorithm based on the repeating purchase behaviors.The author then establishes a precision marketing support system based on data analysis and assesses its marketing effect.The following are the main research results of this article.First,the research proposes a user segmentation model based on improved RFM model and K-means++ algorithm.The extraction of user characteristics uses an improved RFM model to fit the stability characteristics of T-APP's user data,including the user groups,their repeating consumption on certain goods.In this process,the author standardizes the five indexes through the forward and reverse standardization method.The weight of each index is calculated using the entropy method.Then aiming at the shortcomings of K-means,Kmeans++ algorithm is used to analyze user value and improved the accuracy of user segmentation.Second,the research establishes a users' repeat purchase behaviors prediction model based on machine learning algorithms.By analyzing the distribution of customer's purchase behaviors data,the author uses the SMOTE-ENN method to solve the problem caused by unbalanced samples.Then the author uses the TPE optimization algorithm to optimize the hyperparameters of Random Forest,XGBoost,and Light GBM to avoid the complicated parameter optimization process.The soft voting method is adopted then to fuse Random Forest and Light GBM into a RF-Light GBM integrated model which presents a more precise prediction result and a better F1 index comparing to the origin RF,Light GBM?LSTM?CNN-LSTM model as showed in the paper.Third,the research proposes an improved recommendation algorithm based on users' repeating purchase behaviors.Based on the historical consumption data of T-APP,the author excavates the user's repeating purchase behaviors on the community-based e-commerce platform,and presents a recommendation algorithm based on repeating purchase behaviors.Specifically,the author uses mathematical modeling methods to categorize the platform's users into four types,active users with stable interests,inactive users with stable interests,active users with unstable interests and Inactive users with unstable interests.Then the author presents each user category with one specific recommendation algorithm.Namely,the recommendation algorithm for stable interest of active users,the recommendation algorithm for unstable interest of active users,the recommendation algorithm for inactive users' stable interest and the recommendation algorithm for inactive users' unstable interest.Then the author compares the four recommendation algorithms CF algorithm,SVD algorithm,SVD++ algorithm and NMF algorithm.The results show that the improved algorithm is more reliable and has a better performance in the accuracy rate,recall rate and F comprehensive index.Fourth,the paper proposes a decision support system based on data analysis.The research cleans and preprocesses the original data from T-APP platform,based on the results,the author verifies the applicability of SVD++,User?CF and Item?CF recommendation algorithms.The example confirms the applicability and superiority of the improved recommendation algorithm proposed in this paper.Fifth,the research assesses the effectiveness of precision marketing based on an improved AISAS model.Based on the actual situation of T-APP,the author divides the user's behavior into different phases,and re-establishes an indicator system based on AISAS.The result of the calculation using the practical marketing data verifies the effectiveness of the evaluation model.The main contributions of this paper are as follows:(1)Proposes a new way to extract the characteristics of users based on an improved RFM model,its idea of using K-means++ algorithm to analyze the user's value increases the precision of user categorization;(2)Establishes an integrated prediction model to cope with the issues caused by the unbalanced samples and the hyperparameters optimization,thus increases the accuracy of predictions on users' repeating purchase behaviors;(3)Introduces the tempol reward and punishment factor and the cycle of repeating purchases in this research,proposes a recommendation algorithm based on users' repeating purchase interest dynamic,verifies the effectiveness of the recommendation algorithm.