Recommendation Algorithm Of Neural Network And Ensemble Learning Based On User Behavior Sequence

Recommendation system has already been deployed in various fields of network life.In normal interaction scenarios,the usage data generated by users in real time is very sparse.Due to the inflexible model building,the traditional recommendation algorithm does not dig out the relationship between the user's personal preference and the items they browse in the past.In order to fully capture the potential relationship between users and items,the thesis uses the classic Word2 vec method to mine the information in the sequence of user behavior.It is not only applied to the item sequence representing users' preferences,but also innovatively applied to the user sequence representing the audience of items,so as to obtain primary and dense users and item vectors.Combine more attribute features related to the two,and then use two machine learning models MLP and XGBoost to extract deeper nonlinear relationships between different features.Based on this,the thesis builds four different algorithm models: neural network model based on item sequence(IB-MLP),neural network model based on user sequence(UB-MLP),XGBoost model based on item sequence(IB-XGB),and XGBoost model based on user sequence(UB-XGB).The prediction results of these algorithms are verified in the public data commonly used in three recommendation fields,and compared with the other three classical models.The final results show that IB-XGB method has higher accuracy and better robustness.