The Research On Collaborative Filtering Algorithm Based On Item Recommendation System

The rapid development of the Internet has allowed more data to appear around people.The increasing amount of data has greatly facilitated people's lives,and people have got rid of the lack of information to a certain extent.But in the face of such a large amount of data,how to accurately obtain the information people want is a confusing problem.Existing search tools are also powerless in the face of such massive data,and the emergence of recommendation systems has brought the gospel to people's individual needs,and makes information acquisition more efficient and convenient.This paper analyzes several common recommendation algorithms.Although these algorithms are used in different fields,there are still problems such as data sparsity and scalability.In response to these problems,based on the traditional collaborative filtering algorithm,this paper studies the theories of scoring prediction,the collaborative filtering algorithm,similarity calculation and so on,and improves the shortcomings of traditional algorithms finally.Main tasks as follows:(1)Considering the data sparseness problem and scalability problem existing in the traditional collaborative filtering algorithm,and because of different user's scoring scales,it is easy to cause the problem that similar users hard to be found.In the third chapter,through introducing Euclidean distance and dimensionality reduction,a fusion collaborative filtering recommendation algorithm based on improved user similarity and score prediction is proposed.The scoring matrix is improved,and the scoring mean difference is used to calculate the similarity,and the nearest neighbor is found,is ed,and the recommendation is implemented by introducing the average score of the target user and its similar users.Finally,the improved algorithm is combined by the weighted synthesis method,and produces a recommendation resultbased on combing the model of the user's comprehensive interests.The experiment has been carried out several times with the public data set of movielens,and the accuracy and coverage of the recommendation are obviously improved,which proves the validity.(2)For the sparsity problem,this paper based on the similarity calculation method of the popular penalty of the item(User-IIF)and the user(Item-IUF)rejects the traditional method of calculating the sparsity of the number of ungraded matrix units to the total number of cells.considering the influence of the relationship density on the sparsity,defining the new way of calculating the sparsity,and weighting the sparsity,respectively,proposes two cooperative filtering algorithms.It is proved by experiments that the recommended evaluation index MAE value is better than the original algorithm and it has good recommendation effect.