Research On Personalized Recommendation System Based On Logistic Regression Refinement Ranking Model

The rapid development of the Internet makes the amount of network information show an exponential growth trend.This phenomenon brings lots of information to users and also brings information overload problems.It is difficult for users to obtain interesting high-quality information from large amounts of data.In response to this problem,the recommendation system,as an information filtering technology,can filter out valuable information for users from massive information.It solves the redundant problem of Internet information,and improves the efficiency of information utilization.The core of a recommendation system is the recommendation algorithm.Actually,the hybrid recommendation algorithms can achieve better results than the single recommendation algorithms.It introduces a hybrid recommendation method based on content recommendation and collaborative filtering recommendation strategy,which integrates logistic regression refinement ranking model.In addition,crossing features are added to the model to improve the personalization of the recommendation results.The evaluation experiments use user data of a music portal website as the data set.The experimental results prove that our system can effectively improve the recommendation efficiency,and significantly increase the accuracy rate and recall rate.The research includes the construct of recommendation system and the design of recommendation algorithm.The system adopts a three-tier structure,which is divided into a data collection layer,a recommendation kernel layer,and a recommendation system outer layer.This framework can not only perform distributed processing of data,but also has strong scalability of high-dimensional features,and is suitable for large data processing scenarios.The design of recommendation algorithm expounds the establishment of a logistic regression refinement ranking model.It selects three mainstream recommendation algorithms as comparison models,and sets three comparison experiments to prove our method is more universal when both considering the characteristics of users and items.In addition,it propose a method combining crossing features to enhance the personalization degree of the recommendation system,and its effectiveness is proved through experiments.