| The recommender system is to provide a certain reference for people's choices and help them find valuable information.They have become the basic application of e-commerce and information access,effectively pruning a large amount of information space so that users can directly face those projects that best meet their needs and preferences.Collaborative filtering is the most popular method of building recommendation systems and has been successfully applied to many applications.In collaborative filtering,the user's past behavior is analyzed to establish a user-project relationship,to recommend items to the target user based on the opinions of other users,and only to analyze the user's historical behavior data(such as score),usually more than other recommendation systems.Technology has better performance.However,collaborative filtering faces the problem of scoring matrix sparse and cold start.In response to the above problems,the researchers proposed using the matrix decomposition method to build a recommendation system.The matrix decomposition method can effectively solve the highly sparse problem in the recommendation system while ensuring accuracy.In the recommender system,the user expects the system to give a reason for the prediction rather than representing "black box" recommendations.However,the matrix factorization model does not explain the reason for recommendation to the user well.Each factorization dimension of the user and item cannot be explained by the concept in real life,but can only be understood as the latent semantic space.So this study proposes to fuse the user-user neighborhood model with the matrix factorization model and add the time context to the fusion model.The model can be used to solve the problem of data sparsity sensitivity in the user-user neighborhood model and the reason why the user can not explain the recommendation reason in the matrix decomposition model,and improve the prediction accuracy of the recommendation system,which can provide users with a good experience.The main work of this article is as follows:1.The background of the recommendation system was studied,and then the traditional recommendation algorithms were introduced: content-based recommendationalgorithm,item-based collaborative filtering algorithm,user-based collaborative filtering algorithm,matrix factorization model,and study of the principles of these algorithms.These algorithms are described in detail.And compare the advantages and disadvantages of these algorithms:Collaborative filtering algorithm has a good reason for recommendation,can improve the user's experience,beyond the level of improvement can achieve accuracy,its disadvantage is the existence of cold start and data sparsity problems;The matrix factorization model has high scalablity and the accuracy of the algorithm,which can mitigate the problem of data sparsity.However,the matrix factorization algorithm does not know how to explain its recommendation to the user.2.Due to the rapid increase using of the Internet,the recommended project is updated quickly,causing the item-item neighborhood model to be unstable.Therefore,the user-user model and the matrix factorization model are merged to ease the data sparseness in the collaborative filtering algorithm and improve the quality of recommendations.Experiments are carried out to verify that the proposed model's recommendation accuracy is higher than that of the traditional model.And in the movie Lens-100 k data sparsity is very high,it can also be a good recommendation.3.It is undoubtedly that time context is one of the most important contexts,user preferences will change over time,and the popularity of items will change over time.Therefore,this paper proposes that integrating the time context into the improved model can improve the accuracy of the recommendation system and perform experiments on the Movielens database.Comparing with the traditional recommendation system method,the accuracy of the algorithm has been Significantly Improved. |
PDF Full Text Request