Research On Collaborative Filtering Algorithm Combining Scoring,item Attributes And Clustering

With the gradual growth of network interconnection technology,people have entered a period of less information to a period of a lot of information.At this time of huge data volume,if you want to find your favorite content from a lot of information,This is undoubtedly difficult,and these online contents are becoming increasingly difficult to show to people who may like them.In order to solve this problem,the recommendation application came into being,and its mission is to connect everyone and online content,find the connection between them,so as to make personalized recommendations.As a popular recommendation algorithm nowadays: Collaborative filtering recommendation algorithm(CollaboratIve Filtering,CF),its main function is prediction and recommendation.The implementation principle of the recommendation method is to discover what he may like based on the data that the network user has generated on the network,and divide the users on the network into different groups according to the differences in the content that everyone likes and recommend items with similar preferences.CF algorithm is usually divided into two different ways: user-based collaborative filtering algorithm(User-Based CollaboratIve Filtering,UserCF)and item-based collaborative filtering algorithm(Item-Based Collaborative Filtering,ItemCF).Generally,people can be classified according to their preferences,and items can be classified according to categories.This article conducts research based on ItemCF,the main work done is as follows:(1)In order to solve the problems caused by the sparse data of the current CF algorithm and the addition of new users or new projects,this paper designs a CF method that combines the scaling factor and the unique attributes of the project itself.Use the attributes of the added project itself to reduce the problems caused by data sparseness,and then use the ranking of popular items to push for network users to eliminate the push problem when the project(user)is newly added.(2)Due to the traditional ItemCF algorithm,when the number of items is large,the algorithm takes a long time,so according to the operation principle of ItemCF,on the CF algorithm that combines the scale factor and item attributes,a recommendation method combining clustering is designed.By reducing the time it takes to calculate the similarity,the algorithm's processing speed is basically improved without reducing the accuracy of the algorithm.(3)The algorithm uses a more classic data set: Jester and MovieLens and Book-Crossings and other data sources for testing.MovieLens contains many users on the network scoring information for multiple movies,as well as some information about the movie itself such as directors,actors,etc.Information and network scoring user's own information;Jester recommendation system data is captured from Jester Online Joke Recommender System,which is the user's rating data for jokes;Book-Crossings is a book rating data set designed based on the data of bookcrossing.com Mainly contains the user's scoring information on books.The recommended method designed in this article is compared with the traditional CF recommendation method and the method of combining user or project tags to verify the effectiveness of the improved CF recommendation method.