Research On E-commerce Product Recommendation Algorithm Based On Integrated Learning Model

In modern time,different users have individualized requirements.How to mine the user's interest preferences in massive data,so as to make efficient and accurate personalized recommendations has always been the goal pursued by recommendation algorithms.However,since the user's personalized information is difficult to obtain and the user's interests will also migrate in real time,personalized accurate recommendation under the big data environment still faces severe challenges.This article first introduces the preliminary knowledge of the recommendation system,and then introduces the related theoretical foundations required for subsequent model establishment,namely:Extreme Gradient Boosting Tree(XGBoost),Lightweight Efficient Gradient Boosting Tree(LightGBM),and based on Facebook papers.The improved model XGBoost+LR of the intermediate gradient boosting tree and logistic regression regression model(GBDT+LR).Then,based on the data set of user clicks on the Internet data platform,analyze the characteristic factors that may affect the user's click behavior.Through comparative experiments with different data levels,different missing value filling methods,and different models,the model is obtained.The evaluation index reached the optimal data set size,missing value filling method,and model category.The innovation of this article is:firstly convert the click-through rate prediction of the product into the user's interest degree prediction of the product,so that the prediction result is related to the user's behavior;and perform the personalized filling based on the similarity of the behavior of the lack of some user and product attribute information After comparison experiments,this method can further improve the performance of the model;multiple statistical features are added in feature processing,and statistical features occupy the most importance in the ordering of the importance of the optimal model;at the end of the model,the model is constructed On the other hand,the improved gradient boosting tree and the improved XGBoost+LR model based on GBDT+LR are used to predict the user's interest,and users are recommended based on their interest in each candidate item.Finally,after evaluation and analysis of the model results,it is concluded that based on 100,00 click records,the missing values in the data are personalized and filled into the XGBoost model for training,and the optimal recommendation results will be obtained.