Recommendation Algorithm And Optimization Based On Ensemble Learning

Faced with the different needs of users,it is difficult to make accurate recommendation for users in massive data.Personalized recommendation method is a typical strategy to solve the problem.It accurately depicts user portraits based on user's historical behavior data,analyses each user's real preferences and needs,and provides personalized intelligent services for users by using recommendation mechanism.However,due to personalized information is not easy to obtain and user interest transfer exists,personalized recommendation still faces great challenges.In this paper,we find that the personalized information of users has a great impact on the recommendation performance.Full use of feature selection method can effectively extract personalized information.We respectively propose a random forest recommendation model based on feature optimization and an Adaboost recommendation model based on feature enumeration.In the case of user interest transfer,it is found that implicit feedback is affected by potential user interest.Therefore,this paper put forward a personalized recommendation model that integrates multi-angle features.The model fully extracts the effective features and mines the implicit feedback information of users,which improves the precision of recommendation.The main research contents and contributions are as follows:(1)Aiming at the problem that it is difficult to extract effective features due to the large number of user features,this paper proposes a random forest recommendation algorithm based on feature optimization,which integrates classification algorithm into the recommendation process.The model optimizes the number of features by introducing Gini index,and predicts users 'interests and preferences with random forest classification algorithm,which effectively improves the precision and generalization ability of the model.(2)Considering the problem that the feature group is too single to accurately depict the user's personalized preferences,a personalized recommendation model based on ReliefF feature optimization is proposed.This model takes full account of users 'historical behavior and context information,extracts effective features by using ReliefF algorithm,constructs multi-group features by introducing feature enumeration strategy,effectively mines users' personalized preferences,and improves recommendation performance by combining Bayes model and Adaboost algorithm.(3)To solve the problem that personalized information is not easy to obtain and user interest transfer exists,a personalized recommendation model based on Xgboost is proposed,which combines multi-angle features.The model combines the features of users,products,environment and other perspectives,takes full account of the potential preferences and implicit feedback information of users,and predicts user preferences with Xgboost algorithm.The model improves the precision of recommendation.In this paper,the proposed recommendation method is analyzed theoretically and validated experimentally based on the relevant data in the recommendation field,which shows that the proposed methods are effective.