Research On Prediction Of User Purchase Behavior Based On Multi-Index Mixed Model

In recent years,with the development of information technology,the number of products and services that people can access is very large.The main task of the recommendation prediction model is to associate users and information to maximize the value.Traditional prediction algorithm can not implement personalized recommendation for users,which makes users have a great deal of resistance to relevant recommendations.With the continuous development of precision marketing,only the historical consumption data of users is used to construct the users' attribute features,its attribute dimension is too single to predict the potential purchase behavior of users.At the same time,the overload history behavior data has some problems,such as sparse data,missing value and cold start.The user model is developed according to the user's preferences and interests.It can help users find the interested items in a large amount of commodity information.In order to improve the accuracy of the prediction model,a new feature extraction method is used in the process of training set construction,and a hybrid machine learning system is constructed as the prediction model.The main research contents are as follow:(1)User preference feature extraction method based on clustering algorithm and associated RFM model.Most of the previous prediction methods are based on simple rule prediction,which fails to mine the potential association rules,resulting in poor prediction effect.In order to build a complete training set,this paper constructs a training set based on user information,commodity information,user commodity association information and user association information.The main task is to extract features from users' multi-source heterogeneous data.First,the RFM model is used to find valuable users.Then the clustering algorithm and association rules are used to solve the problem of data imbalance and mine more potential related features.Finally,a multi index scoring system is constructed to quantitatively analyze the single behavior data and mixed cross behavior data to form a complete training set.(2)A multi-index hybrid prediction model based on ensemble learning.Previous studies have shown that most machine learning models have certain limitations in feature processing and feature screening,which leads to deficiencies in the efficiency and prediction accuracy of the models.First,we construct a strong classifier with multiple weak classifiers by ensemble learning,and form a scalable tree enhancement system to train the initial feature set.Then,through feature reconstruction,new features and original features are spliced to get a new training set.The tree enhancement system and logistic regression model LR(logistic regression)are fused to construct a hybrid machine learning system to train new feature sets.Compared with the existing schemes,the integrated decision tree model can use fewer resources to train more sample sets.The accuracy is relatively higher than that of the single model,and the prediction effect is better.