Research On Collaborative Filtering Recommendation Algorithms For Data Sparsity

With the rapid development of the Internet and e-commerce,the rapid expansion of information on the network,there has been "information overload" phenomenon.Personalized recommendation technology can help users quickly and accurately from the messy information to find the information needed by users,to some extent alleviate the "information overload" problem.As one of the most widely used personalized recommendation technologies,collaborative filtering technology has achieved considerable success in practical application,but because the actual data are often very sparse,resulting in collaborative filtering technology sparseness of data.The cold start problem is the extreme case of data sparsity problem.This paper studies it as sparseness of data.The sparseness of data seriously affects the recommended quality of collaborative filtering recommendation algorithm.The problem of sparseness of data is due to the increasing number of users and the number of items in the recommended system.The number of users is very small,so the user score matrix must be very sparse,and the cooperative filtering algorithm is very dependent on the user score matrix.In order to solve the sparseness of data,the researchers proposed a number of methods for user scoring matrix,mainly divided into two categories: The first category fills the scoring matrix to reduce its sparseness;The second category is to decompose the scoring matrix and remove those users and projects that have little effect on the computational similarity,which reduces the score matrix dimension.As the second type of method in the deletion of projects and users will inevitably be some useful information is also deleted,so this paper to solve the sparseness of data is used in the first method.The main work of this paper is as follows:1)In order to solve the problem of user's cold start,this paper proposes UserItem-Mix CF algorithm.The traditional cooperative filtering algorithm does not consider the relationship between the items when calculating the similarity between users,which leads to the inaccurate user's similarity.Based on this problem,this paper proposes a method of calculating the similarity between users-Item-Based User Sim.Aims at improving the accuracy of similarity calculations between users;Then,based on the improved user-to-user similarity algorithm,the User-Item-Mix CF algorithm is proposed by adding the user characteristic attribute to the user similarity and fusing the relationship with the item by dynamic balance weight ?.Finally,the User-Item-Mix CF algorithm is compared with the number method in the Movie Lens data set.The experimental results show that the average absolute error(MAE)of the User-Item-Mix CF algorithm are less than the number method when selecting different number of new users.2)In order to solve the data sparseness problem,this paper proposes User-SP CF algorithm.The similarity between users is calculated by using the Item-Based User Sim algorithm,and the calculated similarity between users is filled into the unrated items in the score matrix,which reduces the sparsity of the matrix;In the filled score matrix,find the nearest neighbor set of the target project and complete the recommendation.Finally,on the Movie Lens data set,the User-SP CF algorithm is compared with the collaborative filtering algorithm based on the item score prediction and the item-based collaborative filtering algorithm.Experimental results show: when selecting different number of neighbors,the MAE value of the User-SP CF algorithm is less than the other two algorithms.