Research On Personalized Recommender System Based On Learning To Rank

With the further development of the Internet,The amount of information in the network has increased rapidly.Failure to effectively process and exploit such complex and large information may lead to information overload.The recommendation system can find the connection between users and information from the massive data,and effectively solve the information overload problem.This paper uses learning to rank to reorder the results of different recommendation algorithms.LTR can transform recommendation problem into the sorting problem and improve the recommendation accuracy effectively.Meanwhile,a personalized recommendation system which is designed for the real-time problem was built by using Storm framework.First of all,based on the analysis of real business scenarios,the features that used for individualization recommendation are filtered by using the statistical methods and the method of machine learning such as Gradient Boosting Decision Tree.Meanwhile,in order to enhance the model's ability of expression and forecast precision,the features after screening are combined and discretized.Secondly,the model of Logistic Regression and the model of Factorization Machine are established to study the effects of different feature combinations and parameters on the model.Using the same data set,the differences between the two is compared to pick out the more suitable model.Finally,a personalized recommendation system is established using the Storm Streaming Framework.The system that is designed for the real-time problem is divided into five key modules:reordering module,feature storage module,real-time sample generation module,model storage module and online training module.Reordering module,feature storage module and model storage module are responsible for producing real-time recommendations.Real-time sample generation module and online training module are responsible for the real-time training and updating of the model.This system can provide real-time recommendation services while updating the model,so as to improve recommendation accuracy.